AMD's 12-core "Magny-Cours" Opteron 6174 vs. Intel's 6-core Xeon

[onethreehill]

If the Westmere Xeon EP were a car engine, it would've been made by Porsche. With "only" six cores, each core in the new Xeon offers almost twice the performance of the competition. A 32nm CPU that only occupies 248 mm2 the Westmere Xeon EP embodies pure refinement and intelligent performance, both Porsche traits. It's just made in Portland, not Zuffenhausen.

AMD's offering today is very different. Magny-cours is the CPU version of the American muscle car. It's a brutally large 12-core CPU: two dies, each measuring 346mm2 connected by a massive 24 link Hyper Transport pipe. AMD's Magny-cours Opteron has almost two billion transistors and 19.6MB of cache on-die.
http://anandtech.com/show/2978/amd-s-12-core-magny-cours-opteron-6174-vs-intel-s-6-core-xeon

AMD Opteron 6174 vs Intel Xeon X5650
http://www.bit-tech.net/hardware/cpus/2010/03/31/amd-opteron-6174-vs-intel-xeon-x5650-review/1

[Clairvoyant129]

Forgot to add,


The bottom line is: is this twelve-core Opteron a good deal? For users waiting to use it in a workstation we have our doubts. You’ll benefit from the extra cores when rendering complex scenes, but in all other scenarios (quick simple rendering, modeling) the higher clocked and higher IPC Xeon X5600 series is simply the better choice.

[JumpingJack]

:) This is going to be a fun/interesting thread to read tomorrow.

[Clairvoyant129]

:) This is going to be a fun/interesting thread to read tomorrow.

Yea, love the reactions and overly emotional replies.

[zalbard]

The scores may have something to do with the way Windows is managing the threads.

http://images.anandtech.com/graphs/amd12core_032610044429/22221.png

[acicula]

review was a bit short, overall though amd doesnt get owned as badly as you'd expect.

also 266USD for 8C @2Ghz (AMD Opteron 6128) is pretty darn cheap.

[Stukov]

Forgot to add,

You know that was rather cherry picked and bolded part of the conclusion. If you read the whole article and the full conclusion you see that in some scenarios (as outlined in your quote) Intel was a better choice, but in other scenarios AMD was a better system.

Each had their strengths and weaknesses and the final conclusion was
All in all, this is good news for the IT professional that is a hardware enthusiast. Profiling your application and matching it to the right server CPU pays off and that is exactly what set us apart from the average IT professional.

Which meant when you are choosing which system it is best to you, you need to really look at the kind of loads you will be putting on your server.

[[XC] gomeler]

That went over better than I was expecting. More or less you need to pick the CPU that works best with the workload you are targeting. Looks like another round of HPC contracts for AMD though. Look forward to more benchmarks and reviews and most importantly the prices of the complete servers.

[flippin_waffles]

Nice to see a review up. Though it almost seemed like Johan was a bit annoyed with having to do it.

[wuttz]

some advantages over the competition.

http://i.zdnet.com/blogs/amd032910d.jpg?tag=col1;post-32381

http://i.zdnet.com/blogs/amd032910c.jpg?tag=col1;post-32381


and the biggest advantage for going amd.

http://i.zdnet.com/blogs/amd032910f.jpg?tag=col1;post-32381


http://blogs.zdnet.com/BTL/?p=32381

[kl0012]

A mixed bag of results from a new Opterons. Better then expected in rendering (if this linux Blender alpha version is mature enough to take those results seriously) and a bit worse then expected in virtualization. Still overall a very good results for a new opterons. Yet it has 2x cores then Xeons and 2.5x total die size for virtually the same performance but who cares while it propertly priced (with the exception of AMD itself of cause).

...the Nehalem EX that Intel is going to launch tomorrow...
Interesting. I'm looking forward for the review.

[ajaidev]

lol Anandtech used a X5670 instead of X5660 which is suppose to be the direct competition for the Opteron 6174 according to their own SKU chart... :)

Opteron 6174 - $1165
Opteron 6176 - $1386
X5660 - $1219
X5670 - $1440

So a $300 more expensive and 0.13 Ghz faster cpu, not exactly going by the charts are we.

As for the review AMD's performance is quite good and at the price points quite a marvel. Were the Opteron 6174 cant compete with the X5670 its a close race but where the 6174 does better it does a lot better like this:

http://images.anandtech.com/graphs/amd12core_032610044429/22156.png

Forgot to add in HPC app's AMD really did shine and pulls ahead of X5670 "$1440" in almost all app's very good show indeed. New HPC will have this baby installed thats for sure :)

[wuttz]

A mixed bag of results from a new Opterons...

if you handpick the benchmarks to debut a product in a so-called "review," sure you get a mixed bag.

and i doubt you yourself can afford a beckton, just sayin. :D

[zir_blazer]

The scores may have something to do with the way Windows is managing the threads.

http://images.anandtech.com/graphs/amd12core_032610044429/22221.png
Agree. You can add as many Cores as you want, but not only that by design you are going to be restricted by the law of dimishing returns, but you also need Software to scale as good as possible. If Windows doesn't manage Threads well enough or the Multithreaded applications scales bad with more Threads, then these Opterons aren't going to shine.



gomeler;4312775']That went over better than I was expecting. More or less you need to pick the CPU that works best with the workload you are targeting. Looks like another round of HPC contracts for AMD though. Look forward to more benchmarks and reviews and most importantly the prices of the complete servers.
The issue here is that Xeons fits for more purposes than the new Opterons, and that the true competence for them still has not arrived.


The 6176 looks a bit ridiculous as it delivers only 4% more performance at 30% higher power and 20% higher prices. The real reason behind this CPU is to battle another tanker, the Nehalem EX that Intel is going to launch tomorrow. The TDP and clockspeeds of that huge chip are very similar. If your application scales poorly and you don't care about power consumption, the X5677 is your champion; it is probably the fastest chip on the market for applications with low thread counts.



Nice to see a review up. Though it almost seemed like Johan was a bit annoyed with having to do it.
From the early comments when he described the new Opterons, it made me think that he was quite surprised on the weird design decisions that AMD choose with the platform.
I was unsure about the Hyper Transport Links arrangement until I see the diagram on that Review and was surprised to see that there were some HT Links decoupled from the traditional 16 Bits to 8, makes me remember of the prototype Dual Socket 754 that did the same (It uncoupled the 16 Bits Uncoherent HTL into two 8 Bits, one Uncoherent HTL, and the Coherent HTL for communicating with the other Socket).



Considering the law of diminishing returns, these Opterons scales well from a Hardware standpoint, the most important technical thing that can be learned from them is that the design proved itself to be scalable with decent performance, imposing a new upper limit, showing that it was possible, and in a MCM.
Then there is also an important inherent weakness: A K10 Core can't defeat a Nehalem one, and knowing that Single Threaded performance gives general purposes improvements while adding more Cores depends on application scaling, the Xeons will scale better in most things, while Opterons are application dependant if they are expecting to shine.
I'm not disapointed, because the fact that Nehalems punchs K10 in per Core performance is something that we're used to see, nothing new in that matter. What AMD really needs is a new processing Core. I hope that Bulldozer arrives in time and delivers, as I don't want to face the consequences of informing Hitler the news.

[ajaidev]

A mixed bag of results from a new Opterons. Better then expected in rendering (if this linux Blender alpha version is mature enough to take those results seriously) and a bit worse then expected in virtualization. Still overall a very good results for a new opterons. Yet it has 2x cores then Xeons and 2.5x total die size for virtually the same performance but who cares while it propertly priced (with the exception of AMD itself of cause).

Interesting. I'm looking forward for the review.

http://www.xtremesystems.org/forums/showthread.php?t=247028

[kl0012]

if you handpick the benchmarks to debut a product in a so-called "review," sure you get a mixed bag.

and i doubt you yourself can afford a beckton, just sayin. :D

I can't afford these opterons as well. But my firm can.

[accord99]

As for the review AMD's performance is quite good and at the price points quite a marvel. Were the Opteron 6174 cant compete with the X5670 its a close race but where the 6174 does better it does a lot better like this:
Excluding the conflicting Blender scores, that's the only benchmark where the 6174 does a lot better. And you missed the Oracle Calling Circle bench where the X5670 does a lot better.

[duploxxx]

A mixed bag of results from a new Opterons. Better then expected in rendering (if this linux Blender alpha version is mature enough to take those results seriously) and a bit worse then expected in virtualization. Still overall a very good results for a new opterons. Yet it has 2x cores then Xeons and 2.5x total die size for virtually the same performance but who cares while it propertly priced (with the exception of AMD itself of cause).

Interesting. I'm looking forward for the review.

there won't be any review, since they don't have them, or intel must have dropped a very personal bag to someone holding reviewers hands, it's paper launched and only available from June on.


Forgot to add,


Yea, love the reactions and overly emotional replies.

yeah and you already started off the wrong way.... with your movie type cut and paste.

[ajaidev]

Excluding the conflicting Blender scores, that's the only benchmark where the 6174 does a lot better. And you missed the Oracle Calling Circle bench where the X5670 does a lot better.


As we noted in our previous article, we work with a relatively small database. The result is that the benchmark doesn't scale well beyond 16 cores

2* 6174 = 24cores

24-16 = 8 unused cores...

[accord99]

2* 6174 = 24cores

24-16 = 8 unused cores...
But it still get's beaten by only a 6C/12T Gulftown.

[tool_462]

there won't be any review, since they don't have them, or intel must have dropped a very personal bag to someone holding reviewers hands, it's paper launched and only available from June on.







I think any delay in Beckton reviews is due to lack of boards, chips have been around for 6 months as usual.

[ajaidev]

But it still get's beaten by only a 6C/12T Gulftown.

Dude 16 cores are only used that means if you stick in a pair of Opteron 6134 "8 cores @ 2.3Ghz for $523" they will score higher than the pair of Opteron 6174 "12 cores @ 2.2Ghz for $1165"

In Intels case lets see total nu of cores is 12c and total number of threads is 24t. The program will use 16 threads that means all the real cores will be used as well as 4 fake ones.

Its a win win situation for 6C/12T Xeon


I think any delay in Beckton reviews is due to lack of boards, chips have been around for 6 months as usual.



I have one thing but not the other still waiting for them to call me up so i can take the delivery hand 2 hand :)

[accord99]

Dude 16 cores are only used that means if you stick in a pair of Opteron 6134 "8 cores @ 2.3Ghz for $523" they will score higher than the pair of Opteron 6174 "12 cores @ 2.2Ghz for $1165"
What, maybe 5% higher due to clock speed.

[JF-AMD]

Here is an interesting snapshot.

This is NOT top bin, we were just picking the closest point to Intel's top part from a SPEC INT perspective (you might recall that Integer was their strong point)

http://www.amd.com/PublishingImages/Restricted/Graphic_ChartsDiagrams/BenchmarkJPEG/SPECint_2P.jpg

Same performance, lower power (our standard power to their 130W) and they are 42% higher in price.

[duploxxx]

Here is an interesting snapshot.

This is NOT top bin, we were just picking the closest point to Intel's top part from a SPEC INT perspective (you might recall that Integer was their strong point)

Same performance, lower power (our standard power to their 130W) and they are 42% higher in price.

any official spec-fp results?

[wuttz]

any official spec-fp results?

http://blogs.amd.com/work/wp-content/uploads/2010/03/image0021.jpg



Shifting gears to floating point performance, here we see that we are 21% faster than our competitor’s top 2P product (and again, this is not AMD’s fastest processor.) Best of all we are a whopping 119% faster than our 6-core previous generation processors (formerly codenamed “Istanbul”).


http://blogs.amd.com/work/2010/03/28/welcome-to-the-world-of-12-cores/

[ajaidev]

Here is an interesting snapshot.

This is NOT top bin, we were just picking the closest point to Intel's top part from a SPEC INT perspective (you might recall that Integer was their strong point)

[IMG]http://www.amd.com/PublishingImages/Restricted/Graphic_ChartsDiagrams/BenchmarkJPEG/SPECint_2P.jpg[/IG]

Same performance, lower power (our standard power to their 130W) and they are 42% higher in price.

ammm but the Opteron 6136 is the top bin 8 core part is it not?? Also why not put in Opteron 6168 instead i think its a better cpu for the price than the Opteron 6136 in multi thread situations like above?

[wuttz]

ammm but the Opteron 6136 is the top bin 8 core part is it not?? Also why not put in Opteron 6168 instead i think its a better cpu for the price than the Opteron 6136 in multi thread situations like above?


You may notice that I chose to compete with one speed lower than our top performance processor because we don’t need to top speed to win. We are not only higher performing than their top speed part, but we are 88% faster than our previous generation processors.

You’re probably wondering why we wouldn’t put our highest speed processor up in this comparison. It’s because we realize that while performance is important, it is not the most important factor in server decisions. In most cases, we believe price and power consumption play a far larger role.

http://blogs.amd.com/work/2010/03/28/welcome-to-the-world-of-12-cores/

[zir_blazer]

Well, tomorrow you will have to give your competence a warm welcome. That means that you may have to update that graphic with whatever Intel releases, and if they decide to release the new Processors they may rearrange price accordinly.
While is obvious that you are not going to stand still in your efforts to be a better option than Intel, the K10 Core is holding back all your efforts. Bulldozer NEEDS to put AMD at least at par with current Nehalems if not one step above and have enough headroom to scale to be competitive with Nehalem successor. I still don't know what the hell AMD did bewthem the K8 release 2003 and Barcelona arrival on 2008, they missed an entire generation worth of major Core improvements and lost the tremendous momentum than the K8 brutally owning Prescotts gave them.

[ajaidev]

http://blogs.amd.com/work/2010/03/28/welcome-to-the-world-of-12-cores/

They used a fastest aka top bin 8 core cpu "Opteron 6136 2.4Ghz" and second fastest 12 core cpu "Opteron 6174 2.2Ghz"

If they would have used a Opteron 6168 instead of Opteron 6136 it may have scaled better given the fact it has 4 more cores but is 500Mhz slower. Since both are at the same price i would have loved to see how Opteron 6168 scores against Opteron 6136 most likely it would have been better if not by a lot.

[dartaz]

Some Quotes from anandtech


For users waiting to use it in a workstation we have our doubts. You’ll benefit from the extra cores when rendering complex scenes, but in all other scenarios (quick simple rendering, modeling) the higher clocked and higher IPC Xeon X5600 series is simply the better choice


And then there is the most important segment: the virtualization market. We estimate that the new Opteron 6174 is about 20% slower than the Xeon 5670 in virtualized servers with very high VM counts.

[wuttz]

Some Quotes from anandtech

ummm lal shimpi what!!?
thats intel marketing $$$ at work.
well at least they aren't bribing oem's anymore, afaik...

it renders amazingly fast under linux w/ blender- he has a chart of it, and yet comes to that conclusion. smells funny as rotten fish doesn't it. :ROTF:

http://images.anandtech.com/graphs/amd12core_032610044429/22221.png

[dartaz]

some advantages over the competition.


The pictures you posted was made by AMD

Of course, AMD will always show the advantages, but not the disadvantages. And sometimes they exaggerate the advantages.

[Movieman]

Would anyone here be interested in a direct comparision between an Intel based X5680 system and a dual Magny-Cours 24 core system done by someone you hopefully trust? Me.
I have the Intel system up and running and will have the AMD cpu's shortly, board to follow that and I have all else needed except the heatsinks.
Still wondering what HS are used for the AMD system.

Now what did I get out of that article: They are close in many areas but unless my still opening eyes weren't working I saw that if you can keep a load on that AMD system it starts to show it's strength.
I am VERY interested to see what it will do when I drop an extended full load on all cores in WCG..:D

[duploxxx]

Well, tomorrow you will have to give your competence a warm welcome. That means that you may have to update that graphic with whatever Intel releases, and if they decide to release the new Processors they may rearrange price accordinly.
While is obvious that you are not going to stand still in your efforts to be a better option than Intel, the K10 Core is holding back all your efforts. Bulldozer NEEDS to put AMD at least at par with current Nehalems if not one step above and have enough headroom to scale to be competitive with Nehalem successor. I still don't know what the hell AMD did bewthem the K8 release 2003 and Barcelona arrival on 2008, they missed an entire generation worth of major Core improvements and lost the tremendous momentum than the K8 brutally owning Prescotts gave them.

that chart is for 2p platform so you will only be able to compare the cpu's that are introduced for that, For intel beckton part that will be, I wonder how they will be priced against gulf, not to mention that CPU price might be ok from a msrp point of view but what will be the additional board cost:

E6510 1.73 12 2x4.8 105 4 8
E6540 2.00 18 2x6.4 105 6 12

[Movieman]

that chart is for 2p platform so you will only be able to compare the cpu's that are introduced for that, For intel beckton part that will be, I wonder how they will be priced against gulf, not to mention that CPU price might be ok from a msrp point of view but what will be the additional board cost:

E6510 1.73 12 2x4.8 105 4 8
E6540 2.00 18 2x6.4 105 6 12

Will someone call duploxxx a ambulance, stick some smelling salts under his nose and speak in a gentle volce to him as he regains conciousness from reading than I'm doing a AMD system?:rofl:

[dartaz]

ummm lal shimpi what!!?
thats intel marketing $$$ at work.


What about the AMD fans here in this thread who only show the advantages that AMD have ?

Also, why you are attempting to make it personal ?

[ajaidev]

Would anyone here be interested in a direct comparision between an Intel based X5680 system and a dual Magny-Cours 24 core system done by someone you hopefully trust? Me.
I have the Intel system up and running and will have the AMD cpu's shortly, board to follow that and I have all else needed except the heatsinks.
Still wondering what HS are used for the AMD system.

Now what did I get out of that article: They are close in many areas but unless my still opening eyes weren't working I saw that if you can keep a load on that AMD system it starts to show it's strength.
I am VERY interested to see what it will do when I drop an extended full load on all cores in WCG..:D

It has happened Movieman has been transformed from Blue to Green oh god save us :D

BTW which models are you getting??

EDIT: The X5680 is really out of 12 cores oppys league :). But then again its for $1663, i do hope that you will get Opteron 6176 SE that will make the comparison quite interesting :)

[Movieman]

What about the AMD fans here in this thread who only show the advantages that AMD have ?

Also, why you are attempting to make personal ?

Hello my friend,
Lets not argue today.
We have wonderfull new systems available from both companies and thats something to be happy about yes?
Now matter which someone picks it sure beats the living crap out of the slide rules I used in high school!
If you see a comment that ticks you off look at the screen and say "FU Troll" and get on with life..
It does work..:D

It has happened Movieman has been transformed from Blue to Green oh god save us :D

BTW which models are you getting??

I'll know tomorrow when they arrive.. I think the people responsible see the little kid that still is in me and want to make it a surprise..
Like Christmas in March!:rofl:

[wuttz]

What about the AMD fans here in this thread who only show the advantages that AMD have ?

Also, why you are attempting to make it personal ?

apologies, none of my comments are against you if you read it again.
they were all directed at the quality of his "review" and the most likely intel marketing $$$ he is receiving- hence his skewed reviews.

at anyrate, movieman will be doing an end-user review. i want to see that 2P 12-cores at 100% load with a killawatt on the wall. :up:

[Movieman]

apologies, none of my comments are against you if you read it again.
they were all directed at the quality of his "review" and the most likely intel marketing $$$ he is receiving- hence his skewed reviews.

at anyrate, movieman will be doing an end-user review. i want to see that 2P 12-cores at 100% load with a killawatt on the wall. :up:

and you will.. I bought one just 3 weeks ago.:up:

[N19h7m4r3]

I haven't seen you play with AMD in a while Dave.

I can honestly I'm really excited for your review :D

Hopefully you'll get far more points in WCG than on the W555 :D

[Movieman]

I haven't seen you play with AMD in a while Dave.

I can honestly I'm really excited for your review :D

Hopefully you'll get far more points in WCG than on the W555 :D

I did the AMD 940 a while back and that was a very nice cpu.
This will be interesting, a LONG time since I did a AMD dualie.
I was reading that article and trying to look "between the lines" and it does sound like it's got some real balls.
REALLY looking forward to this.:up:

[ajaidev]

HPC bench's are quite nice for AMD, in estimation one would need X5680 or W5680 to match/pass the score.

http://images.anandtech.com/graphs/amd12core_032610044429/22223.png
http://images.anandtech.com/graphs/amd12core_032610044429/22222.png
http://images.anandtech.com/graphs/amd12core_032610044429/22225.png


I did the AMD 940 a while back and that was a very nice cpu.
This will be interesting, a LONG time since I did a AMD dualie.
I was reading that article and trying to look "between the lines" and it does sound like it's got some real balls.
REALLY looking forward to this.:up:

It has 12 ballz lol, anandtech says its a American muscle car "mustang" and the 6 cores Xeons are porsche "911" :)

[Movieman]

HPC bench's are quite nice for AMD, in estimation one would need X5680 or W5680 to match/pass the score.
It has 12 ballz lol, anandtech says its a American muscle car "mustang" and the 6 cores Xeons are porsche "911" :)
I've got the X5680's so will test at default 3330mhz and depending on which 12 cores arrive try my best to make it a fair honest comparision.

[RaV[666]]

Well, its clearly not a workstation cpu, so giving it light workload isnt going to show its strenghts.Workstation is going to be the other socket, C32 if i recall correctly.This cpu is intended for large raytrace projects and not a small robot render :/.
Anand is comparing it to the more expensive model too, 300$ more.I just cant think of the reason they did it.
By that logic, why they dont compare it to the 300$ LESS expensive intel model.
With a raytrace tests theyve should do some massive scenes that are many GB's in size, like the ones that are done on render farms.

[Movieman]

;4312960']Well, its clearly not a workstation cpu, so giving it light workload isnt going to show its strenghts.Workstation is going to be the other socket, C32 if i recall correctly.This cpu is intended for large raytrace projects and not a small robot render :/.
Anand is comparing it to the more expensive model too, 300$ more.I just cant think of the reason they did it.
By that logic, why they dont compare it to the 300$ LESS expensive intel model.
With a raytrace tests theyve should do some massive scenes that are many GB's in size, like the ones that are done on render farms.

I can, it's what they had.
Reality is even the biggest reviewers don't have every cpu on hand but what they could have done IF they'd wanted to was to reduce the multi in bios and given a fairer comparision.

[ajaidev]

I can, it's what they had.
Reality is even the biggest reviewers don't have every cpu on hand but what they could have done IF they'd wanted to was to reduce the multi in bios and given a fairer comparision.

They had pitted AMD's 2435 against X5570 in the past and that was hardly fair since x5550 was the real price/performance competition.

[RaV[666]]

I can, it's what they had.
Reality is even the biggest reviewers don't have every cpu on hand but what they could have done IF they'd wanted to was to reduce the multi in bios and given a fairer comparision.

Well ,but they didnt, in few instances in this review they didnt even did not do they numbers, some results are estimated.In one instance the benchmark was limited to 16 cores,its not a properly done review...A server cpu is hard to test i know.But if they dont have the resources to do it properly, then maybe they should not.

Back on topic however, if you movieman could do more and maybe more directly comparable tests, i would be the first to thank you :).But before you do a test ,make sure it can use up to 24 cores.Some huge scenes on 3dsmax/maya would be welcomed too, maybe some unix numbers and not wholly windows based.Crunching i figure you already have covered ;-).

[duploxxx]

Will someone call duploxxx a ambulance, stick some smelling salts under his nose and speak in a gentle volce to him as he regains conciousness from reading than I'm doing a AMD system?:rofl:

oh I read that don't worry, but what is wrong with my post ????

zir_blazer makes a statement regarding spec_int spec_fp that will we destroyd according to him. I just point him to facts that there are only few cpu's available on the 2p beckton platform and that they are also minor on specs....

[Florinmocanu]

well, to shed a little light on the Blender/Linux bench, it's accurate.

You get much better results working in linux with blender than with windows. I work with blender quite often and i made a dual boot Linux/W7 just for it, so i don't loose performance due to the OS.
Also, a lot of rendering software has problems keeping 24 cores busy. But, from my perspective as a 3d artist, the 24 cores 2p opterons kick ass, in pure rendering they will destroy any 2p intel system.

I worked 2 years ago with 2 systems, a 4p 16 core opteron system (agena core), at 2 ghz and a 2 p Xeon 5355 at 2.66. In rendering the Opteron system was around 50% faster, in day to day usage, modelling etc... intel lead away due to better IPC and clock speed.

Same scenario repeating again.

PS: i can provide a scene in 3dsmax/mental ray for Movieman to test with. I did it for the lab501 guys, who used it in their Gulftown review.

http://translate.google.com/translate?js=y&prev=_t&hl=en&ie=UTF-8&layout=1&eotf=1&u=http%3A%2F%2Flab501.ro%2Fprocesoare-chipseturi%2Fintel-core-i7-980x-veni-vidi-vici%2F9&sl=ro&tl=en

It's pretty demanding, 6 minutes and 40 seconds on a 4ghz 980X, on older systems you can spend 30-60 minutes or even more.

A list of results lower

http://translate.google.com/translate?hl=en&sl=ro&tl=en&u=http%3A%2F%2Fforum.lab501.ro%2Fshowthread.php%3F t%3D615

[Movieman]

oh I read that don't worry, but what is wrong with my post ????

zir_blazer makes a statement regarding spec_int spec_fp that will we destroyd according to him. I just point him to facts that there are only few cpu's available on the 2p beckton platform and that they are also minor on specs....

Nothing wrong with your post.
I was just having a bit of fun with you with the smelling salts thing.
Lets face it, your seen as a total AMD guy and I'm seen as a Intel guy although thats not true.
What I am is for whatever does the most work in the apps I use with total cost and electrical cost factored in.

[Macadamia]

Err... considering Cinema 4D (aka Cinebench) uses Scanline, the more you scale the less efficient it gets especially when the render image size is small. The overhead isn't because the cores are not being used properly but because it takes time to initiate the render thread, give geometry/scene/position data to the thread and such, and Cinebench is ultimately based off time not pixels/second (flawed in a sense for >8 CPU I guess)


How to scale properly in any renderer:
1. Make it slower by adding more detail and such (so that render init won't be a problem). Renderers like Maxwell Render probably won't have a problem with this because it takes so damn long to render. :rofl:
2. Bucket rendering instead of scanline. This probably helps a lot more.
3. Animation. Dang this is the best solution of all. Unlike Cinebench you'll get proper core usage if the render engine/renderfarm manager supports distributed frame rendering, scanline or bucket rendering doesn't matter anymore. Load balancing should be optimal.

But of course what would Johan know? He's the corporate IT guy.

[gallag]

Is virtulisation still the most important thing? Dont see it mentiond much here?

[Florinmocanu]

there's no need for animation rendering, just make a decent complex scene, that doesn't render in 20-30 seconds. The initial loading time of the scene evens out if the render takes a decent amount of time.

[ajaidev]

Is virtulisation still the most important thing? Dont see it mentiond much here?

Cloud Computing :hail:

[JF-AMD]

any official spec-fp results?

Yeah, that one looks even better:

http://www.amd.com/PublishingImages/Restricted/Graphic_ChartsDiagrams/BenchmarkJPEG/SPECfp_2P.jpg

[CntdwnToExtn]

Is virtulisation still the most important thing? Dont see it mentiond much here?

+1

Most important factor for me deciding within' the next few weeks.

[gOJDO]

Well done AMD! I hope they improve the clocks over time.

BTW, guys, do you remeber when AMD marketing clowns were BS-ing about MCM?
http://plaza.fi/s/f/editor/images/X-2008022609302629611.jpg

[ajaidev]

Well done AMD! I hope they improve the clocks over time.

BTW, guys, do you remeber when AMD marketing clowns were BS-ing about MCM?
[imsg]http://plaza.fi/s/f/editor/images/X-2008022609302629611.jpg[/img]

It was more like the way Intel implemented it than MCM it self. As far as a i can recall MCM via FSB was not like by AMD or something like this. :yepp:

EDIT: @JF-AMD We need a nude pic of a 12 core opteron asap "Without IHS" like the one of q6600 posted by gOJDO above.

[JF-AMD]

Well done AMD! I hope they improve the clocks over time.

BTW, guys, do you remeber when AMD marketing clowns were BS-ing about MCM?


We never said MCM was bad, we said UNCONNECTED MCM was bad. And we still stand by that.

Any communicaiton between the dies happend off-package, in the MCH, and it created additional bus traffic. Which is why Intel ditched that design.

Now, at the time, Paul Otellini was very clear that MCM was a great idea because you could get better yields and improve manufacturability. We never disputed this fact. It will be interesting to see if the other guys flip on MCM or stand by their old statements.

As a person who spent a lot of time talking to the press about technology I can tell you that a.) we always referenced unconnected MCM and b.) the press rarely got the subtlety of the statement. But trust me, we knew what was coming and were very careful about how we structured what we said.

Unconnected MCM, especially with and FSB, was a bad idea.

[kl0012]

We never said MCM was bad, we said UNCONNECTED MCM was bad. And we still stand by that.

Any communicaiton between the dies happend off-package, in the MCH, and it created additional bus traffic. Which is why Intel ditched that design.

Now, at the time, Paul Otellini was very clear that MCM was a great idea because you could get better yields and improve manufacturability. We never disputed this fact. It will be interesting to see if the other guys flip on MCM or stand by their old statements.

As a person who spent a lot of time talking to the press about technology I can tell you that a.) we always referenced unconnected MCM and b.) the press rarely got the subtlety of the statement. But trust me, we knew what was coming and were very careful about how we structured what we said.

Unconnected MCM, especially with and FSB, was a bad idea.

I understand that your primary goal is a discreditation of competitor's products and exhibiting their own products in the best light... but that sentence - Any communicaiton between the dies happend off-package, in the MCH - is completely untrue. An intercore communication (or, in other words, cache synchronization) was happen in the package through FSB which supports cache cocherency protocol. Here are some numbers:
http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=2897&p=2
I would like to see cache2cache synchronization latency in these new opterons.

[Mechromancer]

I want to see HKMG and 3.0Ghz 12-Cores from AMD! 2.2Ghz is good, but 3Ghz is better and would crush the competition. Unfortunately it seems Interlagos (Bulldozer) will be the first HKMG chip from AMD. As it will have higher IPC and clocks (likely), there are no worries about AMD's future.

[OhNoes!]

We never said MCM was bad, we said UNCONNECTED MCM was bad. And we still stand by that.

Any communicaiton between the dies happend off-package, in the MCH, and it created additional bus traffic. Which is why Intel ditched that design.

Now, at the time, Paul Otellini was very clear that MCM was a great idea because you could get better yields and improve manufacturability. We never disputed this fact. It will be interesting to see if the other guys flip on MCM or stand by their old statements.

As a person who spent a lot of time talking to the press about technology I can tell you that a.) we always referenced unconnected MCM and b.) the press rarely got the subtlety of the statement. But trust me, we knew what was coming and were very careful about how we structured what we said.

Unconnected MCM, especially with and FSB, was a bad idea.Could you post a link to ANY source that clearly states that? It seems it is (YOU) AMD that has changed it's tune since its been forced to go to MCM to compete with Intel's superior IPC.

It'll have been sad if AMD didn't learn from Intel's mistakes, No? Besides, AMD's architecture is different (No FSB) so your engineers had to do it differently on your platform.

Finally, LGA 775 C2Qs, even with their "bad" design, are still better overall than your current quad core offerings; in power consumpton, performance at the same clocks, and overclockability on air. These are not empty words.

I do hope you guys do well, but stop bashing your competitor's products, when they're still beating your best offerings at per core, and per watt.

Having said that, I like the competition you guys are bringing to Intel, it's good for the industry and for the first time in 4 years, AMD makes a compelling case for consumers to go AMD with their 6/12 core offerings.

[RaV[666]]

snip

Give it a rest man, for one its old news, since then Intel embraced things amd did first like monolithic die, no fsb, big L3 cache and amd as of today embraced MCM design.
And secondly, LGA 775 MCM and the thing amd starts today are REALLY very much different.Read up.In the end it of course is all about performance and performance per watt.

[Serra]

A very good showing. Kudos AMD!

When competition is strong we the consumer are the biggest winner.

[MrMojoZ]

;4313236']Give it a rest man, for one its old news, since then Intel embraced things amd did first like monolithic die, no fsb, big L3 cache and amd as of today embraced MCM design.
And secondly, LGA 775 MCM and the thing amd starts today are REALLY very much different.Read up.In the end it of course is all about performance and performance per watt.

They can't give it a rest, ever. Poor things.

[Dimitriman]

;4313236']Give it a rest man, for one its old news, since then Intel embraced things amd did first like monolithic die, no fsb, big L3 cache and amd as of today embraced MCM design.
And secondly, LGA 775 MCM and the thing amd starts today are REALLY very much different.Read up.In the end it of course is all about performance and performance per watt.

QFT

fanboys who keep picking on this MCM design change by AMD should really drink a warm cup of :wasntme: since Intel (after K8) has copied as many if not more designs created by AMD.

The overwhelming FACT that stands today is that AMD of 2008-2010 has been under new management and has been much more competitive and stronger in its strategy than the overconfident/slow paced AMD of Hector era. I'm a very unbiased person for both companies and I've owned both Intel and AMD over time and although Intel holds key advantages over AMD today, the company they compete with is completely different than of 2003-2007 and the margin of advantage is getting ever fainter. I am certain that competition will only get tougher with the current management and I give kudos for AMD for being able to turn water into wine with the K10 gen.

[OhNoes!]

;4313236']Give it a rest man, for one its old news, since then Intel embraced things amd did first like monolithic die, no fsb, big L3 cache and amd as of today embraced MCM design.
And secondly, LGA 775 MCM and the thing amd starts today are REALLY very much different.Read up.In the end it of course is all about performance and performance per watt.I KNOW they're different. Did you read my post? And as you have just pointed out, it makes no difference one way or the other if the "crippled" fsb system beats the monolithic die system does it? I agree this is old news, and I have no qualms with AMD for implementing MCM. I am only pointing out the irony of marketing gurus talking down processor designs that actually dominated their own offerings and continue to do so even after all these years. I would rather we drop all this fsb talk and move on.

[flippin_waffles]

[edit 2]

removed.

[FatAlbert]

nude pic of 12 core opteron :D:D nude pic of 12 core opteron :D:D nude pic of 12 core opteron :D:D nude pic of 12 core opteron :D:D nude pic of 12 core opteron :D:D nude pic of 12 core opteron :D:D nude pic of 12 core opteron :D:D nude pic of 12 core opteron :D:D nude pic of 12 core opteron :D:D nude pic of 12 core opteron :D:D nude pic of 12 core opteron :D:D nude pic of 12 core opteron :D:D nude pic of 12 core opteron :D:D nude pic of 12 core opteron :D:D

if not :slapass::horse:

jk :rolleyes::p:

[Movieman]

I'm not so sure. I find it impossible to believe they haven't had the need to obtain a power meter to measure actual power numbers. What kind of server review ignores power from the wall, when it is one of the most important considerations for the product?

The entire article is filled with bias, and it's clear to see the intel slant throughout it regardless of how well the Opteron performed. Opinions may vary! :D

I can't speak to whether there was bias or not on what they did but I can guarantee you there will be none in mine.
Just plain numbers from what I see with my own eyes.

[flippin_waffles]

I can't speak to whether there was bias or not on what they did but I can guarantee you there will be none in mine.
Just plain numbers from what I see with my own eyes.


Looking forward to it. Not sure what kind of server benchmarks you'll be able to do, but it'll be interesting nonetheless! :)

And after thinking about it, I think i'll remove my previous comment.

[onex]

did no one noticed that AMD is running 2.2 vs the 2.93 Intel gave?
that is a killer show!
AMD really did pull out a rabbit from they're hat!

[RaV[666]]

I am only pointing out the irony of marketing gurus talking down processor designs that actually dominated their own offerings and continue to do so even after all these years.

You do know that these old mcm designs are no longer being at top ? IN pure power nehalem architecture dominates, as you would put it, and in power/power cons/price nehalems exchange blows with opterons depending on the workload type.
In server world intel fsb bottlenecked MCMs werent that stellar.On the desktop side was another story.


I would rather we drop all this fsb talk and move on.

Thats the idea :up:

[Movieman]

did no one noticed that AMD is running 2.2 vs the 2.93 Intel gave?
that is a killer show!
AMD really did pull out a rabbit from they're hat!
Think this way: 2200x24=52,800 total MHz
......................2930x12=35,160 total MHz
Then factor in the effect of hyperthreading with the intel.
Like I said, a very interesting build.

[onex]

all amazed with AMD, u lose some 20% probably over HT (rough estimate),

..
that's why asked you to run the POV-Ray bench physical only :yepp:..

[mstp2009]

Not taking sides on this one (I like seeing the competition), but from a manufacturing standpoint I'm sure it is cheaper for Intel in this fight.

248mm^2 vs 2 x 346mm^2.

That's a big silicon difference for basically (on AVERAGE) the same performance.




But kudos to AMD, you gotta work with what you got.

[Movieman]

all amazed with AMD, u lose some 20% probably over HT (rough estimate),

..
that's why asked you to run the POV-Ray bench physical only :yepp:..

You talking to me?
If so please use the quote button so we know who your talking to..:D

[Dresdenboy]

Here is an interesting snapshot.

This is NOT top bin, we were just picking the closest point to Intel's top part from a SPEC INT perspective (you might recall that Integer was their strong point)

http://www.amd.com/PublishingImages/Restricted/Graphic_ChartsDiagrams/BenchmarkJPEG/SPECint_2P.jpg

Same performance, lower power (our standard power to their 130W) and they are 42% higher in price.

Did someone already compare an interesting pair of CPUs out of these charts - the 2.3GHz Barcelona and the MC at 2.2GHz.
SpecInt: 386 vs 105 (3.7x)
SpecFP: 313 vs 95 (3.3x)

24 vs. 8 cores (3x)

So the improvements over the last years seem to fight Amdahl's "law" (based on theoretical reasoning) rather well.

[kl0012]

So the improvements over the last years seem to fight Amdahl's "law" (based on theoretical reasoning) rather well.
I'd say that this is rather PGI compiler improvements (especially in some compiler trick used for peak measurements of 462.libquantum) then a real hardware architectural improvements.

http://www.spec.org/cpu2006/results/res2008q3/cpu2006-20080623-04661.html
http://www.spec.org/cpu2006/results/res2009q2/cpu2006-20090511-07357.html

Also SpecInt_rate benchmark has nothing to do with Amdahl's law.

[Dimitriman]

AMD's offering today is very different. Magny-cours is the CPU version of the American muscle car. It's a brutally large 12-core CPU: two dies, each measuring 346mm2 connected by a massive 24 link Hyper Transport pipe. AMD's Magny-cours Opteron has almost two billion transistors and 19.6MB of cache on-die.

This Muscle car uses a hybrid engine and consumes 20% less petrol than the Porsche...

right?:p:

[onex]

Not taking sides on this one (I like seeing the competition), but from a manufacturing standpoint I'm sure it is cheaper for Intel in this fight.

248mm^2 vs 2 x 346mm^2.

That's a big silicon difference for basically (on AVERAGE) the same performance.

should probably leave AMD with the math, if they can deliver this 1/2 the price of the 74xx and ~the same ability as the 56xx it's probably worthy for them,
that can poses a severe blow to Intel forcing them to sane they're products,
that could mean a huge gain for customers :hrhr:..


You talking to me?
If so please use the quote button so we know who your talking to..
yeah (doh :D) if this post is under you'rs it could generally mean that :yepp:.


that's why asked you to run the POV-Ray bench physical only .. :yepp:

[JF-AMD]

OneX is right. This launch should have a significant impact for customers. That is what this really boils down to. The economics in the server business are changing.

[Manicdan]

Not taking sides on this one (I like seeing the competition), but from a manufacturing standpoint I'm sure it is cheaper for Intel in this fight.

248mm^2 vs 2 x 346mm^2.

That's a big silicon difference for basically (on AVERAGE) the same performance.




But kudos to AMD, you gotta work with what you got.

it is a massive size difference, but as u can see, a lower clock, and much lower voltage 45nm chip for that size, is what lets it compete so well. for servers its all about perf/watt when u have 24 cores to play with. and yes amd may loose some money with the silicon selling for less per mm2, but also how much money do they have to recoup vs intel? i dont even want to think about the difference in research costs between the companies, thats gotta be worse than rocket surgery.

[mstp2009]

OneX is right. This launch should have a significant impact for customers. That is what this really boils down to. The economics in the server business are changing.

As someone who buys a good number of high-end servers, straight up I have to tell you that these products look about equal to me. And the fact that I can drop Gainestowns in existing servers to get 24 threads in one machine is probably the tipping point.

And no, I do not buy the high-end stuff, ever. It's all about quick deployment for a reasonable cost.


If AMD had put this out in Socket F, I would have been all over MC like yesterday. :D

[gOJDO]

We never said MCM was bad, we said UNCONNECTED MCM was bad. And we still stand by that.As far as I remember you(AMD) never said about which MCM you were bit*hing on Intel. But I had enough of hearing about the so called "true"/native/non-glued/etc. quad-core. Ironically, the glued Core2 Quad (the one which the clown Henri was making fun of) whipped the floor with the so called non-glued Phenom. And with a FSB @333MHz(1333Mhz QDR) it had faster inter-core communication between the cores of the different package than the K10(Agena/Barcelona) @2.4GHz.


Any communicaiton between the dies happend off-package, in the MCH, and it created additional bus traffic. For the 1P Core2 based systems, the FSB was more than enough to handle all the traffic between the packages and the rest of the hardware resources. Memory disambiguation and large and fast shared L2 cache? ;)


Which is why Intel ditched that design.Nope. It was because FSB sucked hard in multi-CPU systems, regardless if the CPUs where native dual-cores or MCM quad-cores. Intel needed a new CPU interconnection inorder to compete in the MP server segment.


As a person who spent a lot of time talking to the press about technology I can tell you that a.) we always referenced unconnected MCM and b.) the press rarely got the subtlety of the statement. But trust me, we knew what was coming and were very careful about how we structured what we said. a) I was never aware of that. You(AMD) never specified about which MCM you(AMD) were BS-ing around.
b) I agree that you(AMD) were so careful when you(AMD) started BS-ing us about K10. Henri was carefully BS-ing us about the native design and the dramatic benefits of it, while Randy was BS-ing us about the 40% better performance in various workloads than C2Q(he carefully avoided to specify vs which one).

Prior to Core2 launch, I remember in an interview Herni said: when you have no competitive product you talk about your future. And that's what exactly happened. Henri, Randy and some other guys didn't stopped talking about how great native design is and how fast K10 is going to be. At the end, it not only failed in delivering the promised(or was it a lie on purpose) performance, but it was buggy also (say hello to Cray).


Unconnected MCM, especially with and FSB, was a bad idea.It was a bad idea on Pentium 4, but worked fantastic on Core2.

Anyway, that is now history. I think you(AMD) have learned the lesson about that. Or maybe it is that Hector isn't screwing AMD any more. Never lie to your customers and never make fun of something you are unaware if you'll have to use it to beat the competition.

One more time, you have done a great job with the 12-core Magny and it's MCM approach. I wish you beat Intel on the MP server front this round and earn some money. You'll need it to bring us Bulldozer faster. ;)

[Manicdan]

^wow that was so 2008, i thought weve moved on

[god_43]

Hello my friend,
Lets not argue today.
We have wonderfull new systems available from both companies and thats something to be happy about yes?
Now matter which someone picks it sure beats the living crap out of the slide rules I used in high school!
If you see a comment that ticks you off look at the screen and say "FU Troll" and get on with life..
It does work..:D


I'll know tomorrow when they arrive.. I think the people responsible see the little kid that still is in me and want to make it a surprise..
Like Christmas in March!:rofl:


MM.....that is awesome! although i was just starting to feel confident about catching up to you in WCG.....i can see now that it will never happen ;9.

unless i start up a farm or something.

[mstp2009]

^wow that was so 2008, i thought weve moved on

Well, it was JF-AMD that did bring it up.

[Manicdan]

Well, it was JF-AMD that did bring it up.

it was a reply to guess who:


Well done AMD! I hope they improve the clocks over time.

BTW, guys, do you remeber when AMD marketing clowns were BS-ing about MCM?

[Movieman]

Fed Ex came today.. And I almost did when I opened the package!:rofl:
http://img145.imageshack.us/img145/2856/img0322y.jpg (http://img145.imageshack.us/i/img0322y.jpg/)

http://img338.imageshack.us/img338/2525/img0323v.jpg (http://img338.imageshack.us/i/img0323v.jpg/)

[Agrophel]

Well now as you seen them, Send to me ;D

[Enoc]

Fed Ex came today.. And I almost did when I opened the package!:rofl:

there are adults size pampers you know, better for u than your undies :rofl:

joke apart heeheh , hey movieman, do you have a mobo for the opterons, and what's your plan with them? wcg ppd production comparison vs westmere?

[Movieman]

Well now as you seen them, Send to me ;D
Jeez, you know I would do that but I just don't have a mailing envelope around...NOT!:rofl:

there are adults size pampers you know, better for u than your undies :rofl:

joke apart heeheh , hey movieman, do you have a mobo for the opterons, and what's your plan with them? wcg ppd production comparison vs westmere?

Motherboard, soon..
Plans are build, sit in awe watching, then compare to the X5680 westmere rig and then make the little buggers work their arses off in WCG..:D

and if anyone sees heatsinks for these let me know please..

[Enoc]

and if anyone sees heatsinks for these let me know please..


dynatron builds them...

http://www.dynatron-corp.com/en/product_list.aspx?cv=1-3-289

this seller has them

http://www.acmemicro.com/estore/showproduct.aspx?pid=7902&catid=:lastcatid

http://www.acmemicro.com/estore/showproduct.aspx?pid=7903&catid=384

[Agrophel]

Doh to laite!

make about same noise as GTX 480 :rofl:

http://www.dynatron-corp.com/en/product_detail_1.aspx?cv=1-3-289&id=212&in=0



Noctua's NH-U9

Caution: Please note that the cooler isn't compatible with socket G34.

If you're interested in using our coolers on AMD Opteron platforms, please don't hesitate to contact us at sales@noctua.at! Bundles can be customized for system integrators.

http://www.noctua.at/main.php?show=productview&products_id=22&lng=en

[Movieman]

Doh to laite!

make about same noise as GTX 480 :rofl:

http://www.dynatron-corp.com/en/product_detail_1.aspx?cv=1-3-289&id=212&in=0



Noctua's NH-U9

Caution: Please note that the cooler isn't compatible with socket G34.

If you're interested in using our coolers on AMD Opteron platforms, please don't hesitate to contact us at sales@noctua.at! Bundles can be customized for system integrators.

http://www.noctua.at/main.php?show=productview&products_id=22&lng=en

Now my impression of Noctua's products has always been well built and engineered but for us here in the USA very expensive but for this build I might just say the heck with the cost if you guys can come up with a pair of them that will work on a dual G34 board and not bankrupt me in the process.:D

[onethreehill]

AMD Opteron 6174 vs Intel Xeon X5650
http://www.bit-tech.net/hardware/cpus/2010/03/31/amd-opteron-6174-vs-intel-xeon-x5650-review/1

[duploxxx]

As someone who buys a good number of high-end servers, straight up I have to tell you that these products look about equal to me. And the fact that I can drop Gainestowns in existing servers to get 24 threads in one machine is probably the tipping point.

And no, I do not buy the high-end stuff, ever. It's all about quick deployment for a reasonable cost.


If AMD had put this out in Socket F, I would have been all over MC like yesterday. :D

oh come on what is the sense of your whole reply here? Socket F ddr2 was able to bring you from dual core to hex not to mention the same sock from 2003 till 2009, yes with MC they bring a new sock that is comptible for the future towards buldozer etc....

Now your reply is only valid when you bought a q2 2009 platform.... :cool: that is a new cpu order in about 1 year time, i am sure that won't be very TCO minded.


AMD Opteron 6174 vs Intel Xeon X5650
http://www.bit-tech.net/hardware/cpus/2010/03/31/amd-opteron-6174-vs-intel-xeon-x5650-review/1


oh dear bit-tech that is always a pleasure for a good lauch.... they see a result, never try to find out why this could be like that and just post a comment....
use some non scaling applications in the mix and add an overall score to that, there conclusions are a total EPIC failure.

Power consumption
Interestingly, both systems drew an almost identical amount of power from the wall when both idle and under load. This is quite surprising, as historically speaking Xeons have consumed a lot more power than similarly-priced Opterons. All in all this can be only seen as good news for Intel given that the Xeon X5650s were faster than the Opteron 6174s in the majority of the tests.

historically speaking :D:D oh now the nehalem launch isn't any referral anymore where there conclusion was that Xeon killed the whole amd platform in both cpu power and power consumption, now that AMD is back at the same level they play the old ball again...

Conclusion
AMD has used this as a spur to redesign and re-organise its Opteron brand, dropping backwards compatibility with earlier models in order to simplify its range and boost performance
That said, if you look at the big picture AMD still has its work cut out of it in the HPC/server application space as its latest Opteron 6174 lags behind the similarly priced Xeon X5650 in many of these applications. This is particularly true when you consider than the Opteron 6174 is the second fastest Opteron 6000-series CPU, while there are five faster Xeon 5600-series than the Xeon X5650s we tested.

backwards compatibility WTF is he talking about, did someone else write the conclusion
5 SKU ???? i count 3 5660-5670-5680
the reason why the 6174 isn't destroying the xeon in many benchmarks is the lack of core scaling sw... that should be conclusion, just look at steam how it destroys the 5650

[gOJDO]

Fed Ex came today.. And I almost did when I opened the package!:rofl:
http://img145.imageshack.us/img145/2856/img0322y.jpg (http://img145.imageshack.us/i/img0322y.jpg/)

http://img338.imageshack.us/img338/2525/img0323v.jpg (http://img338.imageshack.us/i/img0323v.jpg/)You lucky #!@!#!%!#@

Come on give me one...:nuts:

[accord99]

oh come on what is the sense of your whole reply here? Socket F ddr2 was able to bring you from dual core to hex not to mention the same sock from 2003 till 2009, yes with MC they bring a new sock that is comptible for the future towards buldozer etc....
You mean 2006 and then again, not every Socket F will support Istanbul.


the reason why the 6174 isn't destroying the xeon in many benchmarks is the lack of core scaling sw... that should be conclusion, just look at steam how it destroys the 5650
No, the reason why the 6174 isn't destroying the Xeon is because the Westmere cores are so much more powerful as to have roughly comparable multi-processing throughput even in excellent scaling applications. And for those applications that can't scale to 24 cores, then it's no contest for the Xeon.

Stream is just a memory bandwidth measurement, the Xeons "destroy" the 6174 in memory latency but that doesn't mean anything without application performance.

[FlanK3r]

Movieman: hope, we will see some overclocking and hope, u get some WR at wprime ,-)

[alfaunits]

@duploAMD:
Historically speaking, power consumption mattered less at the time when AMD was more competitive because the top performers weren't giving enough performance.

It was more fun during K8/K10 era (until MC/Nehalems). You had clear winners in each area, it was easy to make a choice.
Now... you have to look at alllll the freakin' little bits, and even then sample to sample can change the power consumption to something different.

What steam performance??? I hope you don't mean the worthless memory streaming performance...

[Movieman]

Movieman: hope, we will see some overclocking and hope, u get some WR at wprime ,-)

We'll see but at this point I don't know whether there is a way to OC them.
Yes,wPrime I always run..:D

[onex]

We'll see but at this point I don't know whether there is a way to OC them.
K10stat?

[duploxxx]

Stream is just a memory bandwidth measurement, the Xeons "destroy" the 6174 in memory latency but that doesn't mean anything without application performance.

oh pls show some benches where the Xeon is destroying MC on memory latency and bandwidth....

and no i am not talking about just steam but spec wise if you would compare the 5650 with 6174 afterall those are in the same price range and yet only the best 56xx is able to keep spec par. afterall according to all intel fans spec rates are so important for the results....


You mean 2006 and then again, not every Socket F will support Istanbul.


No, the reason why the 6174 isn't destroying the Xeon is because the Westmere cores are so much more powerful as to have roughly comparable multi-processing throughput even in excellent scaling applications. And for those applications that can't scale to 24 cores, then it's no contest for the Xeon.


true about the scaling, but then again Intel is at the limit of there GHZ, don't forget they are already on a 32nm.... while AMD was able to bring this on 45nm, perhaps check the becktons at what power they are rated for a given ghz and core count...

I said ddr2 socket F not any socket F


K10stat?

really wondering if this is working, it isn't working my sr5690 of both istanbul cpu's and the new MC... i can only downclock with K10stat


@duploAMD:
Historically speaking, power consumption mattered less at the time when AMD was more competitive because the top performers weren't giving enough performance.

It was more fun during K8/K10 era (until MC/Nehalems). You had clear winners in each area, it was easy to make a choice.
Now... you have to look at alllll the freakin' little bits, and even then sample to sample can change the power consumption to something different.


totally correct about the past alfintelits but that doesn' change the fact that the bittech conclusion is way off in many ways and there review is only half baked.

[onex]

it isn't working my sr5690 of both istanbul cpu's and the new MC... i can only downclock with K10stat
odd what u'r saying:
http://www.xtremesystems.org/forums/showpost.php?p=3995514&postcount=61
http://img248.imageshack.us/img248/3637/cpuz32.jpg

[Chumbucket843]

2* 6174 = 24cores

24-16 = 8 unused cores...

last time i checked blender uses 8 threads. idk about linux version.

[mstp2009]

duploAMD

ROFL



Dude - you need to CHILL out. No wonder you have the popularity of a bad rash. I was trying to provide some perspective as someone that buys and deploys HPC Clouds for lease.

[Movieman]

I would suggest we all take a breath.
Listen, I know we all have the brands that we like more than "the other guys" but think a little.
They both make excellent products, but yes, different.
A Van Gogh and a Rembrant are both incredible works of art but totally different and appeal to different people.
It's the same thing here and I really wish that people could stop to think on this.

[informal]

^^Listen to the wise man with cool looking mustache :D :D :D .

PS He banned Jupiler! So listen to him even more so! :D

[Movieman]

^^Listen to the wise man with cool looking mustache :D :D :D .

PS He banned Jupiler! So listen to him even more so! :D

And Jupiler is a very good friend!
Banned his butt anyway!
I'll teach that fermi mouthing Belgian!:rofl:

( ever seen a pic of him? Think 6'3 and 270" and NOT fat!.)

[gOJDO]

:rofl:@ duploAMD

I can't remember for what reason, but I have that guy on my ingorelist.


I would suggest we all take a breath.
Listen, I know we all have the brands that we like more than "the other guys" but think a little.
They both make excellent products, but yes, different.
A Van Gogh and a Rembrant are both incredible works of art but totally different and appeal to different people.
It's the same thing here and I really wish that people could stop to think on this. I think you can't identify Intel and AMD with Gogh and Rembrant. You can measure CPUs performance, their power requirements, their performance/price ratio, their OC-ing ability, and other objective factors. But you can't do that to a picture. It has a subjective value. So art has nothing common with technology and pictures has nothing common with CPUs.

IMO if someone is loyal to a certain CPU company, he must be stupid or he must have a benfit of it. Both Intel and AMD had good and bad processors at different point of time, for different purposes and in different segment. If you are loyal to one of them then you must have missed some good CPUs from the other.

This round, I see MC as a winner in MP servers and Intel cought with the pants down. :D If I have to buy a kick-ass MP server, AMD MC would be my choice. But if I have to buy a kick-ass desktop then I won't buy AMD PII because Intel Ci7 is a better choice. So no Gogh/Rembrant analogy.

[radaja]

IMO if someone is loyal to a certain CPU company, he must be stupid or he must have a benfit of it. Both Intel and AMD had good and bad processors at different point of time, for different purposes and in different segment. If you are loyal to one of them then you must have missed some good CPUs from the other.

so very true,but be careful with what say because some might accuse you of using abrasive language.i was accused of using abrasive language for saying almost the exact thing.
i said brand loyalty was the stupidest thing,its just hardware.both companies make cpu's that do the same damn thing.sticking to one brand only gets you 50% of the toys.

i think if you stick to one brand for too long you start on a path of unhealthy obsession and start acting like a political pundit and spout out crazy rhetoric to bolster
your position for no reason what so ever except to point out your brand is better.sad thing is most of these kind of people vow to never even use the other brand and
just point out reviews and charts to prove their point.me personally,i like to try things out myself and come to my own conclusions as to whats good and whats not.
also i like to have 100% fun.:D

[Dresdenboy]

so very true,but be careful with what say because some might accuse you of using abrasive language.i was accused of using abrasive language for saying almost the exact thing.
i said brand loyalty was the stupidest thing,its just hardware.both companies make cpu's that do the same damn thing.sticking to one brand only gets you 50% of the toys.

i think if you stick to one brand for too long you start on a path of unhealthy obsession and start acting like a political pundit and spout out crazy rhetoric to bolster
your position for no reason what so ever except to point out your brand is better.sad thing is most of these kind of people vow to never even use the other brand and
just point out reviews and charts to prove their point.me personally,i like to try things out myself and come to my own conclusions as to whats good and whats not.
also i like to have 100% fun.:D

I think, there is some behavioural pattern in many of us. Think about any fans of a football, baseball, soccer or whatever team. They are sometimes more loyal to their team than even to their wife. ;)

Sure, being fan of any team you would nearly always be on the side of a winner...

[OhNoes!]

so very true,but be careful with what say because some might accuse you of using abrasive language.i was accused of using abrasive language for saying almost the exact thing.
i said brand loyalty was the stupidest thing,its just hardware.both companies make cpu's that do the same damn thing.sticking to one brand only gets you 50% of the toys.

i think if you stick to one brand for too long, [even at the expense of considerable performance loss,] you start on a path of unhealthy obsession and start acting like a political pundit and spout out crazy rhetoric to bolster
your position for no reason what so ever except to point out your brand is better.sad thing is most of these kind of people vow to never even use the other brand and
just point out reviews and charts to prove their point.me personally,i like to try things out myself and come to my own conclusions as to whats good and whats not.
also i like to have 100% fun.:DFanboyitis defined! :clap:

[Stukov]

Someone say Fermi? Oh wait wrong thread.

I think it was Movieman..but did that person who has both sets of CPUs get any benchmarking done on the Intel vs AMD systems?

[Macadamia]

I think you can't identify Intel and AMD with Gogh and Rembrant. You can measure CPUs performance, their power requirements, their performance/price ratio, their OC-ing ability, and other objective factors. But you can't do that to a picture. It has a subjective value. So art has nothing common with technology and pictures has nothing common with CPUs.

IMO if someone is loyal to a certain CPU company, he must be stupid or he must have a benfit of it. Both Intel and AMD had good and bad processors at different point of time, for different purposes and in different segment. If you are loyal to one of them then you must have missed some good CPUs from the other.

This round, I see MC as a winner in MP servers and Intel cought with the pants down. :D If I have to buy a kick-ass MP server, AMD MC would be my choice. But if I have to buy a kick-ass desktop then I won't buy AMD PII because Intel Ci7 is a better choice. So no Gogh/Rembrant analogy.

Actually isn't Movieman spot on?

Excluding subjectivity there will always be technical factors in painting that make us (the general public) favor one over another. People don't always like all artists/ genres of artwork, the initial French Impressionists were pretty much a lonely bunch :p:

/Owaitthisconvoisnowheadedsomewherestrangeandmyster ious

Any yep I agree that loyalty is stupid, but go look at the flipside of a few of the _________________ troll posts, and you'd see the same thing happening on Intel's side: people don't need "decent" graphics (at all!), but turbo mode and fast AES are extremely important etc kind of absurdism.

My laptop's gonna be Arrandale because I appreciate what they're doing and what they're delivering on this timeframe, but server wise AMD kinda pulled off a coup here, not because they are always technologically ahead, but rather because they took a step back and tried to re-evaluate what they could do for the server market with their limited capabilities. (600 USD for 16 cores still baffles me, especially if you can hackishly OC it as an enthusiast...)

[OneEng]

Can someone explain why Westmere does Oracle calling circle benchmark so much better than MC?

http://images.anandtech.com/graphs/amd12core_032610044429/22157.png

It seems nonsensical.

[JumpingJack]

Can someone explain why Westmere does Oracle calling circle benchmark so much better than MC?

http://images.anandtech.com/graphs/amd12core_032610044429/22157.png

It seems nonsensical.

Part of the reason, I believe Johan pointed out in his article -- the benchmark does not scale well beyond 16 cores. So AMD is 8 cores short of well threaded, westmere gets 12 full cores, and some 4 more threads distributed over HT ...

MC will really shine in well threaded situations, the fewer threads you spit out with respect to the two platforms, the results will move toward favoring Westmere.

[Florinmocanu]

Part of the reason, I believe Johan pointed out in his article -- the benchmark does not scale well beyond 16 cores. So AMD is 8 cores short of well threaded, westmere gets 12 full cores, and some 4 more threads distributed over HT ...

MC will really shine in well threaded situations, the fewer threads you spit out with respect to the two platforms, the results will move toward favoring Westmere.

but the benchmark is a really bad one.

Look at intel results.

2 x 5570 (8 cores ) is just barely better than 1 x 5670 (6 cores). I don't care how you say it, but this is bad scaling issues. Or, the review is biased and tried to shine a better light on the new hexacores, i cannot explain otherwise that result.


last time i checked blender uses 8 threads. idk about linux version.

Blender since 2.5 supports up to 64 threads.
It has evolved since the 2.49 a lot on the rendering side.

[Macadamia]

The Bit-tech review is even more ridiculous. Lightwave and Terragen 2?


Someone should try out Maxwell Render 2. 32 cores yield 31++x scaling. This, on a rather complex and branchy algorithm (BDPT). Now these people are actually coding :up:

I think they charge licenses per system too, so Atom renderfarms aren't gonna win here. :p:

[JumpingJack]

but the benchmark is a really bad one.

Look at intel results.

2 x 5570 (8 cores ) is just barely better than 1 x 5670 (6 cores). I don't care how you say it, but this is bad scaling issues. Or, the review is biased and tried to shine a better light on the new hexacores, i cannot explain otherwise that result.

You know, it is hard to tell ... I generally agree, I don't think it is a great benchmark, a great benchmark provides actual performance information relative to the intended usage for that workload.

In terms of scaling/Intel result, there is more than just cores to the 5570 vs 5670 that makes it difficult to really analyze that particular result. For example, I don't know what turbo the 5670 might do over the prior 45 nm generation, the 5670 has 50% more cache and is UMA not NUMA, not sure how that would effect this benchmark... so claiming bad scaling or bias is a bit premature without knowing other extenuating factors.

It is certain though, that because of the lower/poor scaling beyond 16 threads, this is not a bench that would be shining on MC.

[kl0012]

Here it comes again: a benchmark in which the favorite CPU does not win, is not a valid benchmark (or the benchmark/reviewer was biased) - a typical fanboy's position.



2 x 5570 (8 cores ) is just barely better than 1 x 5670 (6 cores). I don't care how you say it, but this is bad scaling issues. Or, the review is biased and tried to shine a better light on the new hexacores, i cannot explain otherwise that result.

If you can't explain this, it doesn't mean that the only explanation is that this benchmark is biased. There are many factors which can affect such types of workloads. For example a cpu2cpu cache synchronization overhead. Or low multithreading scalability factor (see Amdahl's law). Or a limitation of NUMA architectures - even in NUMA aware systems for some types of workloads it is not uncommon when a cpu often needs to access a memory which is connected to a neighboring cpu. One of thoese factors (or a combination of these factors) can lead to this result.

[Florinmocanu]

what? Gulftown is nehalem with 50% more cores and cache. Optimizations are really minor, it's nothing like Conroe and Penryn for example, it's just a bigger Nehalem.

Scaling is really bad going from 6 to 8 cores so there is a problem with the results, whatever the reason may be.

[Movieman]

what? Gulftown is nehalem with 50% more cores and cache. Optimizations are really minor, it's nothing like Conroe and Penryn for example, it's just a bigger Nehalem.

Scaling is really bad going from 6 to 8 cores so there is a problem with the results, whatever the reason may be.

Not correct, Nehalem(Bloomfield and Gainestown) is 45nm.
Westmere is 32nm, more efficient, runs with 6 cores a full 10C cooler than the 45nm quads at the same speeds and is the same physical size and same socket.

[Florinmocanu]

Not correct, Nehalem(Bloomfield and Gainestown) is 45nm.
Westmere is 32nm, more efficient, runs with 6 cores a full 10C cooler than the 45nm quads at the same speeds and is the same physical size and same socket.

i meant it in terms of performance, not consumption/process of fabrication. I know those things, but at the end of the day a 3 ghz 6core gulftown can be maximum 50% faster than a 3 ghz 4 core nehalem. Because it's the same core/same architecture.

and when i see such bad scaling going from 6 to 8 cores i can definitely smell a bit of " let's show gulftown in a good light" intention (personal opinion, not a statement). Scaling should be better, since they used exactly the same platform for all Xeons. The clock speed is the same.

Xeon Server 1: ASUS RS700-E6/RS4 barebone
Dual Intel Xeon "Gainestown" X5570 2.93GHz, Dual Intel Xeon “Westmere” X5670 2.93 GHz
ASUS Z8PS-D12-1U
6x4GB (24GB) ECC Registered DDR3-1333
NIC: Intel 82574L PCI-EGBit LAN
PSU: Delta Electronics DPS-770 AB 770W

[Movieman]

i meant it in terms of performance, not consumption/process of fabrication. I know those things, but at the end of the day a 3 ghz 6core gulftown can be maximum 50% faster than a 3 ghz 4 core nehalem. Because it's the same core/same architecture.

How to answer this?
It takes less elec to run it.
Produces less heat at the same speeds even though it's 6 cores vs 4.
Clocks higher on air at "reasonable" temps.
I see it as a win-win..
What more can you ask from it?:D

[Florinmocanu]

How to answer this?
It takes less elec to run it.
Produces less heat at the same speeds even though it's 6 cores vs 4.
Clocks higher on air at "reasonable" temps.
I see it as a win-win..
What more can you ask from it?:D

I know all those things, i'm planning an upgrade to a 6-core myself, i know what i will gain :D.

I meant the fact that in the oracle benchmark, you see really poor scaling going from 6 to 8 cores, it's almost like the 2 extra cores are not there. Which is really strange.

http://images.anandtech.com/graphs/amd12core_032610044429/22157.png

And they say this on that benchmark

As we noted in our previous article, we work with a relatively small database. The result is that the benchmark doesn't scale well beyond 16 cores.

The scaling stops at 6 cores for Intel, not 16, after that it's really bad.

[Movieman]

I know all those things, i'm planning an upgrade to a 6-core myself, i know what i will gain :D.

I meant the fact that in the oracle benchmark, you see really poor scaling going from 6 to 8 cores, it's almost like the 2 extra cores are not there. Which is really strange.

http://images.anandtech.com/graphs/amd12core_032610044429/22157.png

And they say this on that benchmark

As we noted in our previous article, we work with a relatively small database. The result is that the benchmark doesn't scale well beyond 16 cores.

The scaling stops at 6 cores for Intel, not 16, after that it's really bad.

There are no 8 core chips on that chart. Beckton is 8 core and also 45nm.
The top two chips on that chart are the X5670(6 core) and X5570(4 core)
Where do you see 8 core chips?

Also and don't take this personally, I have a running dual 6 core system for months now..
Currently running dual westmere X5680's at over 4000 on air 100% loaded 24/7 doing DC work..
They greatly out perform the previous 4 core Gainestowns.

I think the issue is your using a benchmark thats limited to 16 threads.
Get one that will see and use 24 threads and then you will see the difference.

[Florinmocanu]

There are no 8 core chips on that chart. Beckton is 8 core and also 45nm.
The top two chips on that chart are the X5670(6 core) and X5570(4 core)
Where do you see 8 core chips?

Also and don't take this personally, I have a running dual 6 core system for months now..
Currently running dual westmere X5680's at over 4000 on air 100% loaded 24/7 doing DC work..
They greatly out perform the previous 4 core Gainestowns.

I think the issue is your using a benchmark thats limited to 16 threads.
Get one that will see and use 24 threads and then you will see the difference.

Look at the chart.

1 x 6 core 5670 vs 2 x 4 cores 5570. = 443 vs 468. That's what i meant.

The talk was about the fact that MC does poorly here, but my opinion is that the benchmark itself is really bad, you see almost no scaling going from 6 to 2x4 (8 cores) and of course a 24 core system won't do that great, the benchmark doesn't scale that well.

[Movieman]

Look at the chart.

1 x 6 core 5670 vs 2 x 4 cores 5570. = 443 vs 468. That's what i meant.

The talk was about the fact that MC does poorly here, but my opinion is that the benchmark itself is really bad, you see almost no scaling going from 6 to 2x4 (8 cores) and of course a 24 core system won't do that great, the benchmark doesn't scale that well.

This we agree on, but I took your orginal comments as against the cpu not the benchmark.
My apologies if I misunderstood.
As soon as I can get my hands on a board I'll see what the MC's can do also.
I have 2-6168's here..Big suckers..

[JF-AMD]

but the benchmark is a really bad one.

Look at intel results.

2 x 5570 (8 cores ) is just barely better than 1 x 5670 (6 cores). I don't care how you say it, but this is bad scaling issues. Or, the review is biased and tried to shine a better light on the new hexacores, i cannot explain otherwise that result.



Yeah, I am not a real fan of benchmarks that don't scale like that. Seems to me that if 8 cores is just barely better than 6 cores, and the benchmark only scales to 16 cores, then the processor with 12 higher clocked cores is going to win.

I have been in this business for almost 20 years now, and outside of the last year or so, I have never seen the oracle calling circle benchmark.

I am guessing that if you ran it on one Magny Cours processor (12 cores) it would probably produce similar results as 2 processors. If that is the case, it would be a clear indication that the workload does not scale and would be a poor indicator for actual performance.

[Movieman]

Yeah, I am not a real fan of benchmarks that don't scale like that. Seems to me that if 8 cores is just barely better than 6 cores, and the benchmark only scales to 16 cores, then the processor with 12 higher clocked cores is going to win.

I have been in this business for almost 20 years now, and outside of the last year or so, I have never seen the oracle calling circle benchmark.

I am guessing that if you ran it on one Magny Cours processor (12 cores) it would probably produce similar results as 2 processors. If that is the case, it would be a clear indication that the workload does not scale and would be a poor indicator for actual performance.

Yes, like using Cinebench 10, just 16 threads max
CB11.5 lets these new chips really show what they have as it supports 24 threads..

[Hornet331]

i meant it in terms of performance, not consumption/process of fabrication. I know those things, but at the end of the day a 3 ghz 6core gulftown can be maximum 50% faster than a 3 ghz 4 core nehalem. Because it's the same core/same architecture.



Oh they add a few things, like ASE-NI, improved virtualization latency new virtualisation capability called "VMX Unrestricted mode support".

Feature wise its roughly the same as it was for the step from conroe to penryn. The reason penryn was bettern then conroe was, that it got more 2nd lvl cache that was even faster then the 2nd lvl cache in conroe.

For westmere intel choose to use the silicon are for more cores and not more cache as the did with penryn.

[Florinmocanu]

This we agree on, but I took your orginal comments as against the cpu not the benchmark.
My apologies if I misunderstood.
As soon as I can get my hands on a board I'll see what the MC's can do also.
I have 2-6168's here..Big suckers..

oh, i'm thrilled about gulftown the same way as i am about magni-cours. Great CPUs. I'm a 3d artist, i render a lot and these CPUs really up the ante in terms of single CPU performance.


But some of these benches are really a poor choice when reviewing, especially considering how highly regarded anandtech is. I was hoping for a bit more self-criticism from them, so that they use only benchmarks which scale well with more then 24 cores.

[Movieman]

oh, i'm thrilled about gulftown the same way as i am about magni-cours. Great CPUs. I'm a 3d artist, i render a lot and these CPUs really up the ante in terms of single CPU performance.


But some of these benches are really a poor choice when reviewing, especially considering how highly regarded anandtech is. I was hoping for a bit more self-criticism from them, so that they use only benchmarks which scale well with more then 24 cores.
Agreed, which is why when I did the X5680 build on the EVGA W555 board I used CB11.5 vs CB10..
To me to do a review and include benchmarks that don't take advantage of all the cores is just plain dumb..
I do include Super Pi but just to give an idea of the relative computational power not as a definitive number.

[JumpingJack]

oh, i'm thrilled about gulftown the same way as i am about magni-cours. Great CPUs. I'm a 3d artist, i render a lot and these CPUs really up the ante in terms of single CPU performance.


But some of these benches are really a poor choice when reviewing, especially considering how highly regarded anandtech is. I was hoping for a bit more self-criticism from them, so that they use only benchmarks which scale well with more then 24 cores.

My opinion is that one can criticize many aspects of that review, nothing really bad, but a few items -- there is a general lack of data, some of the benchmarks are non-industry standard benchmarks (Johan likes to develop his own, real usage type measurements).

The benchmarks are not as much an issue as some of the lack of data, no power measurements for example, some benchmark results are 'estimated' and not actually measured. This is no big deal to me, really, since doing an IT level bench/review is 10x more challenging than doing a simple client review from some tech HW enthusiast site and compounded by the fact that there is limited time with freshly released hardware.

I suspect we will see, as we often see from Johan, several smaller follow up articles where he zeros down on specifics and generates much more data.

Jack

[JF-AMD]

There will always be some criticism of any benchmark, that will never end.

As to the power, the system that he had was not a production system, it was a test system which is not optimized for power. If he ran power benchmarks they would have been very iaccurate, a level far higher than any actual customer would ever see.

Tec Channel had some power benchmarks, the 80W ACP Opteron was under the power of the 95W TDP Xeon. So, for those that argue that ACP is not real and only TDP matters, they would have a hard time explaining how a 115W TDP part (with 80W ACP) could perform under a 95W TDP intel part. The bottom line is that both companies measure TDP differently, which is why power at the wall is the best measurement.

[Hornet331]

There will always be some criticism of any benchmark, that will never end.

As to the power, the system that he had was not a production system, it was a test system which is not optimized for power. If he ran power benchmarks they would have been very iaccurate, a level far higher than any actual customer would ever see.

Tec Channel had some power benchmarks, the 80W ACP Opteron was under the power of the 95W TDP Xeon. So, for those that argue that ACP is not real and only TDP matters, they would have a hard time explaining how a 115W TDP part (with 80W ACP) could perform under a 95W TDP intel part. The bottom line is that both companies measure TDP differently, which is why power at the wall is the best measurement.

Oh you mean the review where they used different PSUs and different kind of rams...?
The only point that the review make is that system A consumes less/more then System B as a whole package, you can't make any assumption about what different component in each system consume.

[kl0012]

what? Gulftown is nehalem with 50% more cores and cache. Optimizations are really minor, it's nothing like Conroe and Penryn for example, it's just a bigger Nehalem.

Scaling is really bad going from 6 to 8 cores so there is a problem with the results, whatever the reason may be.

This is not about "core optimization". Going from 1 socket to 2 socket system you may expect a perfect scaling only if each thread uses its own dataset which is not always possible in case of DB systems. The overhead of access to a non-local memory may eat a big chunk of potential additions to the performance from additional cores.

[gOJDO]

Tec Channel had some power benchmarks, the 80W ACP Opteron was under the power of the 95W TDP Xeon. Like Hornet elaborated, they have compared total system power consumption where all the systems had different RAM types, brand&models and configurations as well as different PSU's. So, their power consumption benchmarks have nothing to do with the CPUs power consumption.


So, for those that argue that ACP is not real and only TDP matters, they would have a hard time explaining how a 115W TDP part (with 80W ACP) could perform under a 95W TDP intel part. As far as I understand ACP and TDP are measuring different parameters. TDP is related to the thermal design, while ACP is related to the power consumption. As for the explanations, the same can be said about Agena's and Yorkfield's. Agena's were producing more heat than Yorkfield's with lower rated TDP.


The bottom line is that both companies measure TDP differently, which is why power at the wall is the best measurement.I agree on this. I'll add: both companies have different TDP measuring methods for different generations of their CPUs.

[alfaunits]

Had you looked more at the scores you would notice the dual X5670 scales quite a bit over single X5670, so that core count argument of single X5670 to a dual X5570 doesn't hold ground. There is definitely somet uArc change that favors the new tech over the "old" one.
Plus the socket count that kl0012 mentioned.
That's how you explain the results ;)

Has anyone seen temperature on these babies, BTW? (both MC and X56xx) I didn't find 'em in any review. (just power consumption)


but the benchmark is a really bad one.
Look at intel results.
2 x 5570 (8 cores ) is just barely better than 1 x 5670 (6 cores). I don't care how you say it, but this is bad scaling issues. Or, the review is biased and tried to shine a better light on the new hexacores, i cannot explain otherwise that result.

[Florinmocanu]

Had you looked more at the scores you would notice the dual X5670 scales quite a bit over single X5670, so that core count argument of single X5670 to a dual X5570 doesn't hold ground. There is definitely somet uArc change that favors the new tech over the "old" one.
Plus the socket count that kl0012 mentioned.
That's how you explain the results ;)

Has anyone seen temperature on these babies, BTW? (both MC and X56xx) I didn't find 'em in any review. (just power consumption)

No. I disagree.

1 x 5670 = 443
2 x 5670 = 580

That equals a 30.9 % increase with double core count.

That's not favoring of newer tech or better results for Intel, that's bad benchmark, which is not usefull to test workstations with 12-16-24 cores/threads.

[JF-AMD]

yeah, that is completely true. If you look at core scalability we are close to 90%. One would imagine the competitor should scale the same way (cores only not on hyperthreads).

If you are only seeing 30% scalability, something is not right. If you have enough physical cores for both you should be much higher.

[alfaunits]

No. I disagree.
1 x 5670 = 443
2 x 5670 = 580
That equals a 30.9 % increase with double core count.
You did not mention that. You said 6 to 8 cores do not scale. I just showed they did. Never said it's a perfect scale...


That's not favoring of newer tech
6 cores of new tech work as fast as 8 cores of last one is not favoring newer tech? How is that?


or better results for Intel, that's bad benchmark, which is not usefull to test workstations with 12-16-24 cores/threads.
Dude, they are not supposed to just show benchmarks that test 24 cores/threads - that's an AMD biased benchmark then ;)
They ARE supposed to show as many benches as possible so you can tell if it makes sense to upgrade and what to upgrade to.

Just comparing tech doesn't sell - you don't buy a CPU because something shows it is superior - but because it is superior at what you need to do.
You also buy the whole system not just the CPU. Not much use of 5000 cores or whatever if the rest of the system cannot follow.

[OneEng]

Thanks guys for all the analysis.

Here is my story so far (on the Oracle Calling Circle delima):


This benchmark does not scale well with number of cores based on the 5570 scores for single and dual socket scores (little difference with doubling of cores).
Something else is making up a HUGE amount of difference in this benchmark for the Westmere (X5670) since 12 X5670 cores scaled much more than simple core scaling from 4-8 cores showed on the X5570.
Magny-Cours 24 cores running at only 2.2Ghz is getting a pretty raw deal in this benchmark as it can't seem to spread its wings and breath (it would be good to see how utilized those 24 cores were ;) ).


So is the Oracle Calling Circle benchmark representative of a major market segment ...... or is it just a bad benchmark for the target markets of the MC processor?

[duploxxx]

Oh they add a few things, like improved virtualization latency new virtualisation capability called "VMX Unrestricted mode support".


cut/paste from some sites or actually know what you are talking about.....you just provided some nice chitchat from marketing.

The real mode / unpaged mode was already available with AMD barcelona while the emulation was still required with Intel until now and it is only required for linux code and only when you enable EPT-NPT.

On the latency intel had to show the improvement after there underperforming VMexit latency, lets say with nehalem and westmere they finally catch-up with amd, last time i found a doc that showed amd shanghai latency was about half the time of a penryn core.



Thanks guys for all the analysis.

Here is my story so far (on the Oracle Calling Circle delima):


This benchmark does not scale well with number of cores based on the 5570 scores for single and dual socket scores (little difference with doubling of cores).
Something else is making up a HUGE amount of difference in this benchmark for the Westmere (X5670) since 12 X5670 cores scaled much more than simple core scaling from 4-8 cores showed on the X5570.
Magny-Cours 24 cores running at only 2.2Ghz is getting a pretty raw deal in this benchmark as it can't seem to spread its wings and breath (it would be good to see how utilized those 24 cores were ;) ).


So is the Oracle Calling Circle benchmark representative of a major market segment ...... or is it just a bad benchmark for the target markets of the MC processor?

from anandtech:

We now use an new even heavier log. As the Nieuws.be application became more popular and more complex, the database has grown and queries have become more complex too. The results are no longer comparable to previous results. They are similar, but much lower.

Nieuws.be MS SQL Server 2008 - New Heavy log!

Pretty amazing performance here. And while AMD gets a pat on the back, it is the hard working people of Microsoft SQL Server team we should send our kudos to. Our calculations show that SQL Server adds about 80% of performance when adding an extra 12 cores, which is simply awesome scaling. The result of this scaling is that for once, you can notice which CPUs have real cores vs. ones that have virtual (Hyper Threading) cores: the 12-core Opteron 6174 outperforms the best Xeon by 20%. The people with transaction databases should go for the Intel CPUs, while the data miners should consider the latest Opteron. The architectures that AMD and Intel have chosen are complete opposites, and the result is that the differences between the different software categories are very dramatic. Profile your software before you make a choice! It has never been so important.

[Florinmocanu]

You did not mention that. You said 6 to 8 cores do not scale. I just showed they did. Never said it's a perfect scale...


6 cores of new tech work as fast as 8 cores of last one is not favoring newer tech? How is that?


Dude, they are not supposed to just show benchmarks that test 24 cores/threads - that's an AMD biased benchmark then ;)
They ARE supposed to show as many benches as possible so you can tell if it makes sense to upgrade and what to upgrade to.

Just comparing tech doesn't sell - you don't buy a CPU because something shows it is superior - but because it is superior at what you need to do.
You also buy the whole system not just the CPU. Not much use of 5000 cores or whatever if the rest of the system cannot follow.

What have you been smoking? The review is testing systems capable of 24 threads and you say it would be AMD biased?

Than let us test a dual gulftown system with super-pi or playing a game, that is surely relevant right?

and going from 443 points for 6 cores to 460-470 for 8 cores is not scaling, not in the real world.

[Hornet331]

cut/paste from some sites or actually know what you are talking about.....you just provided some nice chitchat from marketing.

The real mode / unpaged mode was already available with AMD barcelona while the emulation was still required with Intel until now and it is only required for linux code and only when you enable EPT-NPT.

On the latency intel had to show the improvement after there underperforming VMexit latency, lets say with nehalem and westmere they finally catch-up with amd, last time i found a doc that showed amd shanghai latency was about half the time of a penryn core.



Funny how you bring in amd, when it was not aimed at amd. The question or better the claim was, that there where no architectural changes from nehalem to westmere (aka its the same core).
I just pointed out what changes where made, i can only rely on marketing slides, since thats all, thats available to me. Sorry that i have no super deep inside knowlage what exactly intel modified in westmere... :rolleyes:

Again it was nothing to do with amd at all, it cool that amd has already feature that where recently implemented in westmere, but that still doesn't change the fact, that its a new feature compared to nehalem.

[kl0012]

and going from 443 points for 6 cores to 460-470 for 8 cores is not scaling, not in the real world.

Not every application in the "real world" scales perfectly with the number of cores (even less applications scales well with the number of sockets). Just face it. And "not a perfect scaling" dosn't makes application less "real world". It just not an every algorithm/workload in the world can be effectively parallelized. And even if some algorithm has a good potential for multithreading, not every programmer (or team) can make it effectively. Clock speed still matters.

[alfaunits]

What have you been smoking?
Nothing, I am allergic to both.


The review is testing systems capable of 24 threads and you say it would be AMD biased?
No, it is testing TWO systems, ONE of which is 24 cores. We all know HT can help sometimes, but it's not a core.
If they only test apps that are 24-thread capable, then it is not a test of the two systems - it is a test that shows where AMD can win = AMD biased test.
Is that indicative of ANY other scenario? Of course not.
I don't want them to cover just scenarios that use 24 cores - I want them to cover all apps, so I can see if what I use has any use of the extra cores. If it doesn't, WTF do I care if AMD can run something else better?


Than let us test a dual gulftown system with super-pi or playing a game, that is surely relevant right?
Seriously, if you compare an Oracle benchmark with a game, you're not supposed to post in this thread.


and going from 443 points for 6 cores to 460-470 for 8 cores is not scaling, not in the real world.
Not the same cores, not the same socket number. You're wasting our time repeating the same stuff over and over.

[gOJDO]

What have you been smoking? Chika Kure told me that betaunits is smoking ajduchka trava. And that is some really good $h!t...

[alfaunits]

Yep, my bro betaunits smokes just the top notch stuff ;)

[Movieman]

Hey guys, c'mon, lets get this at least close to being on topic huh..:p:
Here's a thought for you;
Many of you know I have a working dual westmere system on a EVGA SR2 board.. Yes, it is a beast and love it.
I also received a pair of 12 core AMD MC 6168(1900mhz) cpu's and looking forward just as much to that build to see what they can do at default and maybe a tad beyond default..:wasntme:
My point is see things for what they are, an advancement in technology that we all benefit by.
Sure one or the other will be X% better in certain tasks but one thing is sure, both are better at all tasks than the last generation from the same companies cpu's..That is the reason all of us should be smiling.
Think if JUST 3 years ago someone said you'd have 24 core/thread desktops you'd look at him and laugh..
No laughing now and I think we owe both these companies a thanks for what they've made available to all of us.

[alfaunits]

Exactly what I said - one is better at something, the other at something else. Otherwise, one would not have come to market :D
So asking for only 24-core tests is asking for "let AMD win" :)

Let us know the temps of those MCs, I wanna know!

[Movieman]

Exactly what I said - one is better at something, the other at something else. Otherwise, one would not have come to market :D
So asking for only 24-core tests is asking for "let AMD win" :)

Let us know the temps of those MCs, I wanna know!
My gut tells me they will run in the high 30's-low 40's max at 100% load..
Remember, these are the 6168 1900mhz versions not the hot rod :D 6174 2300mhz versions..

[Mescalamba]

I want eVGA SR3 for AMD.. argh!

[alfaunits]

My gut tells me they will run in the high 30's-low 40's max at 100% load..
Remember, these are the 6168 1900mhz versions not the hot rod :D 6174 2300mhz versions..

A 12-core at <40*C?? How many Deltas per CPU as you using? :D

[Movieman]

A 12-core at <40*C?? How many Deltas per CPU as you using? :D

Just one.. the 363CFM ones..:rofl:
No, those numbers are just an educated guess..

[alfaunits]

I know Istanbul is quite cool, but I can't imagine 12 cores being that cool.
Would be nice though... I got highly multithreaded work, a single 12 core could be useful!

[hyc]

Not every application in the "real world" scales perfectly with the number of cores (even less applications scales well with the number of sockets). Just face it. And "not a perfect scaling" dosn't makes application less "real world". It just not an every algorithm/workload in the world can be effectively parallelized. And even if some algorithm has a good potential for multithreading, not every programmer (or team) can make it effectively. Clock speed still matters.

If you're buying many-core hardware, and the software you're trying to run was not effectively parallelized by its programming team, you need to get different software from a better programming team.

[=[PULSAR]=]

Any luck with finding a motherboard of choice yet Dave?

[duploxxx]

Funny how you bring in amd, when it was not aimed at amd. The question or better the claim was, that there where no architectural changes from nehalem to westmere (aka its the same core).
I just pointed out what changes where made, i can only rely on marketing slides, since thats all, thats available to me. Sorry that i have no super deep inside knowlage what exactly intel modified in westmere... :rolleyes:

Again it was nothing to do with amd at all, it cool that amd has already feature that where recently implemented in westmere, but that still doesn't change the fact, that its a new feature compared to nehalem.

true your only mentioning features, but then again this is a MC vs Gulf thread so compare is at the right thread.

[gOJDO]

I know Istanbul is quite coolI agree. There are a lot of beautiful places in Istanbul that have to be seen.:p:


Many of you know I have a working dual westmere system on a EVGA SR2 board..And that's why many of us are jealous to you. :slobber:


Yes, it is a beast and love it.We love you too Dave. We even love your logo. I mean, the new XS logo.


I also received a pair of 12 core AMD MC 6168(1900mhz) cpu's and looking forward just as much to that build to see what they can do at default and maybe a tad beyond default..:wasntme:The bolded part is what intrigues us the most. OC-ing servers must be fun.

So, since you have the mainboard, the two MCs and four Thermi's, how long do you think we'll be patient to wait for some OC-ing results/benches? :rolleyes:

[ajaidev]

i hope he overclocks in to 2.3ghz to simulate AMDs highest end package.

[gOJDO]

If they can release a 2.2GHz now, I think they can release at least a 2.4GHz part by the end of the year. That's the "beauty" of MCM. :D

[FlanK3r]

i think, he can maybe more, Stephan had 3GHz MC :)

[Movieman]

i think, he can maybe more, Stephan had 3GHz MC :)

Listen: For the 12,930,582,016 time, S7's chips were unlocked!
Look at his thread..19 multi''s..PLUUUUUEEESE!:rofl:
The ones I have here are locked at 9.5 and you think I can hit 3GHz with those? Now who's smoking the wacky tobacci?:rofl:
Lets see..uh..3000 divided by 9.5 equals a FSB of 315.8..
Nope, ain't gonna happn cap'n!:p:

i hope he overclocks in to 2.3ghz to simulate AMDs highest end package.

That thought has crossed my mind once or six hundred times..:wasntme:

[FlanK3r]

Movieman: Ok, ok :D. But I am non-smoker and no drugs, i have only protein for fitness :D.
Maybe 240-250x9.5...?

[Movieman]

Movieman: Ok, ok :D. But I am non-smoker and no drugs, i have only protein for fitness :D.
Maybe 240-250x9.5...?

Maybe 240 if we can get around all the little obstacles life tosses at you working on a new platform with absolutely no info around!:rofl:
It ain't easy my friend, figuring out EVERYTHING, every setting,etc..
I remember sitting in a chair for most of 3 days tring to get a C0 965 Nehalem Bloomfield to 4226 on air..Being clueless on all the settings and no one to go to for info..
People forget, I'm a stability guy, not a max clock guy like most here.
Yea, I enjoy playing and seeing what they will do maxxed out but when that 15 mins is over it's back to finding the max safe 24/7 100% loaded speeds.

[gOJDO]

But I am non-smoker and no drugs, i have only protein for fitness :D.Drugs for fitness is just like OC-ing your CPU. You increase its performance, but your reduce its life. :rofl:

@Movieman Stop complaining and find some Cu pots and LN2.:rolleyes: 315MHz base clock is doable!

[SocketMan]

With all these mobo companies out there,is there one with big enough balls
to make an OCing G34 motherboard ? 2P and/or 1P with core unlocking function (in addition to OC).

[RaV[666]]

With all these mobo companies out there,is there one with big enough balls
to make an OCing G34 motherboard ? 2P and/or 1P with core unlocking function (in addition to OC).

Nope, EVGA doesnt do amd stuff unfortunately :/ .Only hope would be with bios modders, but i doubt any of them is going to buy a mobo to try it.
There is possibility that chipset is some kind of 890FX mutation, maybe some low level registers tweaking could be possible.

[demonkevy666]

;4327413']Nope, EVGA doesnt do amd stuff unfortunately :/ .Only hope would be with bios modders, but i doubt any of them is going to buy a mobo to try it.
There is possibility that chipset is some kind of 890FX mutation, maybe some low level registers tweaking could be possible.

evga usually does NV chipsets, the lost a lot of time with 780A to market and by that time it was too late, am3 was around the corner.
they didn't bother making any because they had no experience with am2+ and am3 really.

somewhere is was mentioned evga won't do it until a new architecture.
the new bulldozer architecture will be on G34, and will be the new architecture after Magny-Cours.
someone would have to convince them of this and possible show proof.

[Xanadu]

Not sure if this has been posted yet, but the first G34 board is up on Newegg.

I prefer to wait for Supermicro myself, but it's good to see them coming out in full retail.

http://www.newegg.com/Product/Product.aspx?Item=N82E16813151212

[alfaunits]

$450 only? Holy shiite ;)

[vitchilo]

At least it's cheaper than the SR-2 for Xeons.

Now if only they could have the 8 cores version of the opteron...

My mistake they have them now...

Now I need benchmarks.

[instred]

they do
http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=2010340727%201652756958&name=Socket%20G34

[Movieman]

Not sure if this has been posted yet, but the first G34 board is up on Newegg.

I prefer to wait for Supermicro myself, but it's good to see them coming out in full retail.

http://www.newegg.com/Product/Product.aspx?Item=N82E16813151212
They are ready! Take your pick..single,dual or quad!:p:
http://www.supermicro.com/products/nfo/AMD_G34.cfm?pg=MOBO

$450 only? Holy shiite ;)
SM's projected prices on the duals are $440-470

At least it's cheaper than the SR-2 for Xeons.
.
Yes, but a SR2 is a totally overclockable top end board.
Can I be blunt with you?
The $600.00 EVGA is asking for it is a frigging steal when you look at what your getting for that money.
The one I have is a pre release board and rock stable at 100% load 24/7 and I've taken it to 4500 on air..
Like I said, a steal at $600.00

[Sly Fox]

Yes, but a SR2 is a totally overclockable top end board.
Can I be blunt with you?
The $600.00 EVGA is asking for it is a frigging steal when you look at what your getting for that money.
The one I have is a pre release board and rock stable at 100% load 24/7 and I've taken it to 4500 on air..
Like I said, a steal at $600.00

What would be considered the high-end pricing for other top quality DP boards?

$1000 or so?

Just curious.

[Movieman]

What would be considered the high-end pricing for other top quality DP boards?

$1000 or so?

Just curious.

$600.00 is at the top end for some of the SM top DP boards.
Prices run from $450-600
BUT hard to compare to the EVGA SR2 as it offers overclocking features that no other company does.
Listen everyone knows I'm a big fan of Supermicro boards because of the quality and the company's excellent support of thier products but the EVGA is in a different league than both SM and Tyan in this case.
There is simply nothing to fairly compare it to.
When you go to quad socket boards then you start to move into the $1000.00 range.

[vitchilo]

What would be neat is if someone on this board would make a CUSTOM BIOS for one of those G34 boards to allow overclocking.

I know I'm asking the moon, but I'm sure it's possible.

[Movieman]

What would be neat is if someone on this board would make a CUSTOM BIOS for one of those G34 boards to allow overclocking.

I know I'm asking the moon, but I'm sure it's possible.

It is but also a lot of work. Only two people I know of with the skills and to be honest it's a imposition to ask someone to do that much work without getting them some $$$ for it.
Now get 20 people who want the same board to toss $50 each and that might be doable and cheap for everyone involved.

[richierich]

Wow awesome.

[SocketMan]

I love the color on this one:

http://www.supermicro.com/Aplus/motherboard/Opteron6100/SR56x0/H8QG6-F.cfm


Tyan has a nice lineup as well
http://www.tyan.com/product_board_list.aspx?cpuid=4&socketid=24&chipsetid=99999

[vitchilo]

Those quad sockets boards are E-ATX, smaller than the SR-2? What a joke. EVGA is a joke with their EEEEEEE-ATX board. At least those quad sockets fit in a normal case.

EDIT: on TYAN website it was written E-ATX format but it's in fact a MEB format...

:)

[alfaunits]

The quad sockets are MEB and SWTX, from what I can see. Not EATX.

[Sam_oslo]

I don't know if this has been posted before, but it is better to have tow of these goodies than none.


ASUS Releases KGPE-D16 Socket G34 Motherboard for 12-core AMD Opteron Processors (http://www.techpowerup.com/119540/ASUS_Releases_KGPE-D16_Socket_G34_Motherboard_for_12-core_AMD_Opteron_Processors.html)

http://www.techpowerup.com/img/10-04-07/23a.jpg

[ndrD0g]

It is but also a lot of work. Only two people I know of with the skills and to be honest it's a imposition to ask someone to do that much work without getting them some $$$ for it.
Now get 20 people who want the same board to toss $50 each and that might be doable and cheap for everyone involved.

I would gladly pay $50 to be able to OC these monsters :)

[Sadman]

I don't know if this has been posted before, but it is better to have tow of these goodies than none.


ASUS Releases KGPE-D16 Socket G34 Motherboard for 12-core AMD Opteron Processors (http://www.techpowerup.com/119540/ASUS_Releases_KGPE-D16_Socket_G34_Motherboard_for_12-core_AMD_Opteron_Processors.html)

http://www.techpowerup.com/img/10-04-07/23a.jpg

Looks like SR-2...only for g34...:confused:

[Sly Fox]

$600.00 is at the top end for some of the SM top DP boards.
Prices run from $450-600
BUT hard to compare to the EVGA SR2 as it offers overclocking features that no other company does.
Listen everyone knows I'm a big fan of Supermicro boards because of the quality and the company's excellent support of thier products but the EVGA is in a different league than both SM and Tyan in this case.
There is simply nothing to fairly compare it to.
When you go to quad socket boards then you start to move into the $1000.00 range.

Ahh I gotcha. It's not that the SR2 is cheap in comparison, it's just incredibly high quality.

Thanks for clarifying. :up:

[Movieman]

Ahh I gotcha. It's not that the SR2 is cheap in comparison, it's just incredibly high quality.

Thanks for clarifying. :up:
No, thats the point, it's not comparable to the others and it's not a quality thing, it's the feature set in the bios that sets it apart.
In ALL DP boards except the EVGA SR2 there are no adjustments for vcore,vdimm,VTT,etc..
The cpu's run at default and thats it.memory runs at 1.35 or 1.5v..No ability to get top mamory and use 1.65v..
The EVGA board is like a top gaming single socket board with all the adjustments you see in those single socket boards BIOS.
Think of it as EVGA 762 classified board with one more socket and your close.

[Sly Fox]

No, thats the point, it's not comparable to the others and it's not a quality thing, it's the feature set in the bios that sets it apart.
In ALL DP boards except the EVGA SR2 there are no adjustments for vcore,vdimm,VTT,etc..
The cpu's run at default and thats it.memory runs at 1.35 or 1.5v..No ability to get top mamory and use 1.65v..
The EVGA board is like a top gaming single socket board with all the adjustments you see in those single socket boards BIOS.
Think of it as EVGA 762 classified board with one more socket and your close.

Despite my lazily formed comment I actually did get what you were saying. :rofl:

I should have said high quality bios tweaking features or something instead I think.

In any case, the value of a top-notch OC'ing board for dual chips has been made clear to me now. :D

[Hap]

I really like this Supermicro Motherboard, thinking of buying two.
http://www.supermicro.com/Aplus/motherboard/Opteron6100/SR56x0/H8SGL-F.cfm

[Movieman]

I really like this Supermicro Motherboard, thinking of buying two.
http://www.supermicro.com/Aplus/motherboard/Opteron6100/SR56x0/H8SGL-F.cfm

If those boards had 2-PCI-e X16 slots I might grab one myself.

[Hornet331]

If those boards had 2-PCI-e X16 slots I might grab one myself.

Well... theres always the option to "mod" it. :D

It possible to use a 16 lane card in a 8 lane slot, just cut the end of the slot open and viola.

Though its still only a pci x8 slot, but its fast enough for most stuff. ;)

[Hap]

With a ATI 100-505603 FirePro V8800 card, one slot should be enough

http://www.amd.com/us/Documents/ATI_FirePro_V8800_DataSheet.pdf

[snoro]

It is but also a lot of work. Only two people I know of with the skills and to be honest it's a imposition to ask someone to do that much work without getting them some $$$ for it.
Now get 20 people who want the same board to toss $50 each and that might be doable and cheap for everyone involved.

WAIT!!WHAT???? you have some connection that can make some custom bios and you never told me ??????

I will be back. I have to give a good lesson to MM.

So anyone is up for a group buy on some 4 socket g34 motherboard ???

Hmm 32 core or 48 core at 3ghz. 144k estimated ppd for 48 core at 3ghz. Now we are talking about a true dedicated cruncher and not that little evga sr-2 that can only produce as low as only 100k ppd.:rofl:

[Movieman]

WAIT!!WHAT???? you have some connection that can make some custom bios and you never told me ??????

I will be back. I have to give a good lesson to MM.

So anyone is up for a group buy on some 4 socket g34 motherboard ???

Hmm 32 core or 48 core at 3ghz. 144k estimated ppd for 48 core at 3ghz. Now we are talking about a true dedicated cruncher and not that little evga sr-2 that can only produce as low as only 100k ppd.:rofl:

I guarantee you your not going to get a quad socket anything to clock the way you want.. dual yes,with the right tools but quad no..

[snoro]

What tool are you talking about ? Dual socket could be more doable for me with all my other hobby(mostly stuff for my soccer choaching job and Litterature) eating quite a bit of my paycheck

[mibo]

I'm interested in such tool, too.
Also, if there is the possibility to get somebody hack some oc functionality into a dual or quad socket g34 board, my $50 would be in.

[haylui]

I'm interested in such tool, too.
Also, if there is the possibility to get somebody hack some oc functionality into a dual or quad socket g34 board, my $50 would be in.

I think you need to rise your $50 a few dozen folds before someone would do that...

[mibo]

I think you need to rise your $50 a few dozen folds before someone would do that...

Or, we look for some dozen of people willing to pay for such BIOS.

[onex]

It is but also a lot of work. Only two people I know of with the skills and to be honest it's a imposition to ask someone to do that much work without getting them some $$$ for it.
Now get 20 people who want the same board to toss $50 each and that might be doable and cheap for everyone involved.
that is a great idea, if doable, it might be even possible to migrate the code to different mobo's with only minor changes..?
or maybe supporting K10stat developer to add selected MB to his program?
what do you say?

[Vinas]

higher vmware licensing costs per 6 cores...

[JF-AMD]

higher vmware licensing costs per 6 cores...

That really doesn't hold as much water.

First, there are no core restrictions for ESX 3.x, which is the majority of the VMware installations. So if you are going to make that claim, you have to put the right qualifiers in, you can't say "VMware costs more", you have to say VSphere (which most are not migrating to) has the restrictions.

Secondly, there is not a higher licensing cost, you just need to use the Advanced edition, vs. the standard. You get a lot of redundancy and RAS features on the advanced that you don't get on standard:

http://www.vmware.com/products/vsphere/buy/editions_comparison.html

I did some quick shopping on HP's site. Configured 2 systems as close as I could:

DL380G7, 2 x X5670, 72GB memory, 1 SATA drive, VSphere standard $11,315.
DL385G7, 2 x 6174, 64GB memory, 1 SATA drive, VSphere advanced $11,423.

So, on one hand you get 8 extra GB of memory and a stripped down hypervisor. On the other, you get 2X the cores and a fully featured hypervisor.

If you are building a platform for virtualization, I would recommend the more cores and more robust virtualization.

[JF-AMD]

Dell has finally launched their 4P PowerEdge R815. The value 4Ps are now here. If you thought that the real competition was 2P MC to 2P Westmere, think again.

There are plenty of places that a 4P can give you significantly better performance than a 2P. And if it is priced less, that makes it even more attractive.

http://links.amd.com/Redefining4P

It's not the be-all end-all solution for every workload but there are some serious places that this platform can make a dent in your data center tasks.

[onex]

can't tell which is better,
prices for the MC and this AMD "come back" (can it be said?),
or Intel being able to make in one 6 core chip, what amd is doing with a dual 6 core die one.

yet
there is the stock vs stock clock added to it of course.
the only premium you left paying is to Dell...

anyhow,
thought the 4P market is a counterpart to the becktons,
there is no way comparing 4 2.2 12 core MC's to 2 X5680,
benches showed even an advantage to the opterons at 2P configuration, and even the opterons are/were with a ~1GHz less clock speed.

aside from that,
shame these systems won't be probably commonly used for distributed computing (maybe except from one prize winner?)
for the casual user only the 2P might do, the 4 are still too expensive,
not complaining though.

[skycrane]

Dell has finally launched their 4P PowerEdge R815. The value 4Ps are now here. If you thought that the real competition was 2P MC to 2P Westmere, think again.

There are plenty of places that a 4P can give you significantly better performance than a 2P. And if it is priced less, that makes it even more attractive.

http://links.amd.com/Redefining4P

It's not the be-all end-all solution for every workload but there are some serious places that this platform can make a dent in your data center tasks.

Hey JF, I've got a ? I've got 3 4P tyan 8356 opti rigs crunching, and I'm wondering will the 6core optis work in my mobo?

[JF-AMD]

Depends on the board. There are 2 things that you need, one is dual dynamic power management (split plane) and the other is a BIOS that supports it. Tyan's website should tell you both.

If it has DDPM but does not show support in BIOS, it *may* work, but I generally don't recommend doing that. I know that my Opteron 1389 works in all of my Gigabyte boards, even though it is not in the BIOS, but if I fry everything I know where to get more processors and I am only out $100 for the board. For 6-core 8000 series you have more on the line in terms of cost.

[duploxxx]

That really doesn't hold as much water.

First, there are no core restrictions for ESX 3.x, which is the majority of the VMware installations. So if you are going to make that claim, you have to put the right qualifiers in, you can't say "VMware costs more", you have to say VSphere (which most are not migrating to) has the restrictions.

Secondly, there is not a higher licensing cost, you just need to use the Advanced edition, vs. the standard. You get a lot of redundancy and RAS features on the advanced that you don't get on standard:

http://www.vmware.com/products/vsphere/buy/editions_comparison.html

I did some quick shopping on HP's site. Configured 2 systems as close as I could:

DL380G7, 2 x X5670, 72GB memory, 1 SATA drive, VSphere standard $11,315.
DL385G7, 2 x 6174, 64GB memory, 1 SATA drive, VSphere advanced $11,423.

So, on one hand you get 8 extra GB of memory and a stripped down hypervisor. On the other, you get 2X the cores and a fully featured hypervisor.

If you are building a platform for virtualization, I would recommend the more cores and more robust virtualization.

nobody buys standard anymore, standard and enterprise are still existing because of the old migration path from 3 to 4. These days you use the freeware esxi platform - advanced or ent plus. The essentials are also nice but those are prebuild packages that can't be expanded, so real smb solutions, they don't have the money most of the time anyhow to buy expensive platforms.

On the other hand JF, everybody is migrating to vSphere4 since you always have subscription, so you are allowed to migrate 1 version (from 3 to 4) , the advantages are way to high against version 3 for sure on older/less core solutions due to the enhanced cpu scheduling.

[jwcolby]

I am a business owner on a very real very tight budget. I need as much SQL Server power as I can get but SQL Server licensing discusses sockets not cores, and SQL Server efficiently uses multi-threading. I.e. if I license a single socket then the most threads I can get on that socket is what is important. From my reading SQL Server may or may not efficiently use Intel's hyperthreading. If it does use it efficiently, it still does not scale 2X but rather something like .3X more. If it doesn't use it efficiently it actually hurts efficiency. If it weren't for that fact then it would be a harder decision but it makes no sense to me to have to test what I am doing today to see if hyperthreading is going to help or hurt and turn it on or off. If hyperthreading may hurt, it gets turned off, at which point real cores (and cost / core) are all there is to discuss.

1) Cost per 100% efficient cores matters.
2) Cores per socket matters for Microsoft SQL Server licensing. It costs no more for 16 cores than for two *if* the cores are on the same socket. If they are on different sockets then the license cost is linear as I add sockets.
3) Cores and memory are manually assigned to SQL Server and some of both has to be intentionally left for the OS.
4) Total cost of the mb/cpu/memory matters a great deal.

Obviously if I were a big firm with big IT dollars then I might be considering a high price Intel 10 core processor. On my budget it is a non-starter. On my budget the Magney-Cours makes sense to me.

[HotGore]

I am a business owner on a very real very tight budget. I need as much SQL Server power as I can get but SQL Server licensing discusses sockets not cores, and SQL Server efficiently uses multi-threading. I.e. if I license a single socket then the most threads I can get on that socket is what is important. From my reading SQL Server may or may not efficiently use Intel's hyperthreading. If it does use it efficiently, it still does not scale 2X but rather something like .3X more. If it doesn't use it efficiently it actually hurts efficiency. If it weren't for that fact then it would be a harder decision but it makes no sense to me to have to test what I am doing today to see if hyperthreading is going to help or hurt and turn it on or off. If hyperthreading may hurt, it gets turned off, at which point real cores (and cost / core) are all there is to discuss.

1) Cost per 100% efficient cores matters.
2) Cores per socket matters for Microsoft SQL Server licensing. It costs no more for 16 cores than for two *if* the cores are on the same socket. If they are on different sockets then the license cost is linear as I add sockets.
3) Cores and memory are manually assigned to SQL Server and some of both has to be intentionally left for the OS.
4) Total cost of the mb/cpu/memory matters a great deal.

Obviously if I were a big firm with big IT dollars then I might be considering a high price Intel 10 core processor. On my budget it is a non-starter. On my budget the Magney-Cours makes sense to me.

Good question, you should create a new thread rather than bump this old one.

[Mad Pistol]

Dude... I just read through multiple pages, and didn't even look at the posting date.

great bump...

[erek]

can we get a pic of the Magny-Cours without an IHS please?

[AliG]

lolwut where did that bump come from?

[haylui]

I am a business owner on a very real very tight budget. I need as much SQL Server power as I can get but SQL Server licensing discusses sockets not cores, and SQL Server efficiently uses multi-threading. I.e. if I license a single socket then the most threads I can get on that socket is what is important. From my reading SQL Server may or may not efficiently use Intel's hyperthreading. If it does use it efficiently, it still does not scale 2X but rather something like .3X more. If it doesn't use it efficiently it actually hurts efficiency. If it weren't for that fact then it would be a harder decision but it makes no sense to me to have to test what I am doing today to see if hyperthreading is going to help or hurt and turn it on or off. If hyperthreading may hurt, it gets turned off, at which point real cores (and cost / core) are all there is to discuss.

1) Cost per 100% efficient cores matters.
2) Cores per socket matters for Microsoft SQL Server licensing. It costs no more for 16 cores than for two *if* the cores are on the same socket. If they are on different sockets then the license cost is linear as I add sockets.
3) Cores and memory are manually assigned to SQL Server and some of both has to be intentionally left for the OS.
4) Total cost of the mb/cpu/memory matters a great deal.

Obviously if I were a big firm with big IT dollars then I might be considering a high price Intel 10 core processor. On my budget it is a non-starter. On my budget the Magney-Cours makes sense to me.

then now you can consider Interlargos

[masterg]

stop bumping this thread

[o-brian]

I have my Supermicro H8QG6-F running with 3x AMD 6212 (8Core@2600MHz).
The fourth socket unfortunaly did not work :-(
But the performance of the System is very well.
A performance Test i have not done yet but I will try and put the results here.

Regards, O-Brian

[gallag]

the never ending thread :eek:

[Sparky]

Whoa old thread

[Movieman]

Yes it is Sparky..
Here's some thoughts from someone whose owned both dualie 1366 and the AMD dualie G34
The Intel system has more brute computational power with the top hexes and a lot more BUT the AMD is a very interesting system.
I have two of the lower 12 cores(6168's at 1900MHz) on the Asus KGPE-D16 board.
It is just the most dependable and solid system I have ever owned and I have had this since they first came out in April 2010.
Load it solid to 100% and it just hums along. This summer with the huge heat we've seen it was at times running in a 88F room and still no issues at full load.
Well over two years and no issues. Well, I should say one: When we've lost power totally from ice storm,etc I've had to pull the battery and clear the CMOS but bear in mind that
I've never updated the bios and I'm on the orginal bios released so this may have been corrected.
Also this system should handle VM's better than the Intel but thats just my beleif.
On the MJ12 app I do witch is all network requests the AMD is hands down better than any of the Intel machines here.
Yes, core for core it is weaker than the Intel but it has 24 legit cores to handle the work.
Wish I had a couple of the new 16 cores to see how they work.:D
Yes, I can without any reservations recommend this system to anyone.:up:

[zanzabar]

Yes it is Sparky..
Here's some thoughts from someone whose owned both dualie 1366 and the AMD dualie G34
The Intel system has more brute computational power with the top hexes and a lot more BUT the AMD is a very interesting system.
I have two of the lower 12 cores(6168's at 1900MHz) on the Asus KGPE-D16 board.
It is just the most dependable and solid system I have ever owned and I have had this since they first came out in April 2010.
Load it solid to 100% and it just hums along. This summer with the huge heat we've seen it was at times running in a 88F room and still no issues at full load.
Well over two years and no issues. Well, I should say one: When we've lost power totally from ice storm,etc I've had to pull the battery and clear the CMOS but bear in mind that
I've never updated the bios and I'm on the orginal bios released so this may have been corrected.
Also this system should handle VM's better than the Intel but thats just my beleif.
On the MJ12 app I do witch is all network requests the AMD is hands down better than any of the Intel machines here.
Yes, core for core it is weaker than the Intel but it has 24 legit cores to handle the work.
Wish I had a couple of the new 16 cores to see how they work.:D
Yes, I can without any reservations recommend this system to anyone.:up:

look at the folding and wprime benches, that shows what will happen in a VM (atleast when testing with windows) also look at the performance per watt and per $$$ amd wins all the way. it is the only thing that is amd is useful for right now other than the apu.

[kuroikenshi]

Yes it is Sparky..

Well over two years and no issues. Well, I should say one: When we've lost power totally from iceCREAM storm,etc I've had to pull the battery and clear the CMOS but bear in mind that


I am so sleepy that this is how I read the line and went...:confused:

[SnipingWaste]

Ill have my Magny-Cours 12 core ES up at the end of this month. The company I work for is giving us a single server spot for free. Its a 2U spot in our data center with a IPv4 block /29 with 100mb drop and an IPMI drop. Im waiting for my Supermicro 2U case to be shipped here. Its the one Movieman reviewed with 6 hotswap SATA/SAS drives slots. The ES has no multi lock and I tested it running at 3.6gig on all 12 cores at full load. Im still thinking what Im going to do with it. I have 32gig of ram too so should do VM with out problems.

[M.Beier]

Here's some thoughts from someone whose
And then some from a grumpy youngster;
Its a years old thread, absolutely NO news value in july 2012, if you guys want to discuss the topic, fine, no problemo, but wouldnt a smart admin move, since you actually want to give your advice/help to the person having concerns - be to MOVE THE THREAD to AMD PROCESSOR section, where it would belong now?

Just my 5 cents,