Intel Xeon 5600 Secures, Speeds Cloud Computing

[mstp2009]

Makes we wonder about people overclocking their folding machines. You are getting more work done, but is there some type of error checking that compares results so there is redundancy of data?

Yes, there are multiple levels of error checking.

And the same Work Unit is distributed to multiple machines so there is redundancy checks in the returned data.

[terrace215]

I don't want to rain on their parade, let them have their day in the sun. You'll all know soon enough. I am smiling big time today.

Quick! You should give the details to Danielle Cheisi right now!

Nice game face, though. nVidia is good at the "just you wait!" game, too.

Do let us know when AMD manages to grab, say, 10% of the server MPU market by revenue.

In the meantime, I hear there's an under-development 32nm process with uncontrolled variability that needs fixing. Gate-first.... whoops!

[-Sweeper_]

Quick! You should give the details to Danielle Cheisi right now!

Nice game face, though. nVidia is good at the "just you wait!" game, too.

Do let us know when AMD manages to grab, say, 10% of the server MPU market by revenue.

In the meantime, I hear there's an under-development 32nm process with uncontrolled variability that needs fixing. Gate-first.... whoops!

[Sgt.McRuff]

I wouldn't say I am 100 percent certain, but the fact that it is a quad core with 12 MB of L3 cache, I would have to say a disabled hexa-core CPU.

I would expect that, it's a common practice to improve yields, it makes a lot of financial sense.

Naw I seriously doubt that. Intel prefers to make chips with all their cores functioning, so this simply has to be monolithic quad core.

Intel seems to like saving silicon, but I wonder. I think its disabled hex, but want some hard fact if its disabled hex or quad with extra L3.

[SocketMan]

Good news,too bad it took almost a year after AMD's 6 core Opty release,
we would have had much better prices/choices if it happened a few months ago.Better late then never,I suppose.

As far as the 6 core VS 4 core Xeons - if both have the same number of transistors then obviously the quads have 2 cores disabled,same with
MCs (8 vs 12).Cmon Asus give us a G34 mobo with "core unlocker" or maybe come up with something for the Xeons

[kl0012]

Dual Gulf overclocking...
http://www.theinquirer.net/inquirer/...up-benchmarked

This thing beats 24 opteron cores at 2.8GHz with a good margin

[saaya]

Looking at the Table you posted, the 5620 actually HAS HT and other Features. This one seems to be the most interesting one, if it works on X58MB and is OC'able

yeah, but its almost 400$... but true, if the turbo multi is really that high, then thats ok... i just wonder how well they clock...

lol saaya negative as allways, intel always cuts features in the lowerend xeons, look up the 550x vs the 55x0. And the new 56x0 series is a direct replacement of the 55x0 series.

They new 56x0 have the same price as the old 55x0.

5620=5520
5630=5530

And with the 5650 you get a hexacore for the same price as the old 5550.

The only thing that is a bit different is the 56x7, which is a new product, guess its a slavaged hexacore.

well i didnt dream up 500$ hexa cores and 250$ 32nm quads... there were rumors/leaks mentioning those and the prices were really sweet...

Is NDA is over?

who cares...

Oh no. Just released new line of intel enterprise server chips are really expensive? It must mean yield problems. Yeah, that's it.

a few months ago they werent as crippled and didnt cost that much... there was supposed to be a 600$ hexa core at lower clocks for example, and a 32nm quad for 300$ or less...

It's Intel's decision, but i wish there was a 2.26-2.4 ghz hexa-core at 500-600 USD. I would buy it for sure.

well originally there was supposed to be one... no idea what ever happened to it... im pretty sure that original leak was reliable, hence my conclusion that intel might have worse than expected yields...

@Saaya:
The 5650 is a 21 multi 6 core w/HT and under a grand .
Also to all, keep that L5640 in mind at $774.00
That's a hell of a chip and should do 4000 in a good X58 single socket board.
For the guys that want speed not power the X5677 is what you want.
27 multi, 3600 on turbo and that might run close to 5GHz on air.

under a grand... but if you spend that much you might as well get the 980 and have unlocked multis...

5g on air? i find that hard to believe... intel had some 975 samples that did that as well, but the retail chips never got close to it...
well, for 1500$ it BETTER runs 5g... but even if it does, for that price it really doesnt make sense...

i thought intel would lower prices with 32nm... but all they do is reduce mfc costs and sell them for the same price as 45nm chips and make more money off of them, meh :/

in the end, intel managed to turn their xeon lineup into a pure server chip lineup again and no longer offers chips that might be interesting for overclockers... except for one or two models maybe... so i guess we will have to wait and pray that maybe they will release 32nm desktop quads after all :/

[duploxxx]

Dual Gulf overclocking...
http://www.theinquirer.net/inquirer/...up-benchmarked

This thing beats 24 opteron cores at 2.8GHz with a good margin

yeah right adding ln2 unstable platforms, hardly able to run anything and compare it against a 24/7 platform real blue fanboy.

Can someone run just a dual gulf @ 3ghz and 3.33 do the prime all over again movieman?

[kl0012]

yeah right adding ln2 unstable platforms, hardly able to run anything and compare it against a 24/7 platform real blue fanboy.

Nice try. But read it again - this Gulfy doesn't need 5 GHz to beat 2.8 GHz 24 core opteron (4.1 GHz is enough), which mean that even without overclocking (3.33GHz) it will beat 2.2 ghz magny-cours (especialy when JF confirms that this platform probably wont offer overclocking). Leave alone Nehalem-EX. And since when someone who overclocks his server/workstation system is a "real blue fanboy"? Seem you are on the wrong forum.

[Chris_redfield]

To be fair Intel doesn't really support OC'ing with the Xeon line up. EVGA has done all the work for them, and their's is the only DP 5520 board to do it.

Maybe if we get lucky someone will look at doing an overclockable DP amd board, but lets face it the market for overclockable DP boards and server chips is tiny and probably not profitable at all.

[savantu]

I believe I found the answer why Magny Cours was priced so low, less than 1k for the top high end part. The reasons is more than obvious when you look at the performance : it is bound to suck and cost a lot SW wise. Considering that for a lot of SW your license fee is proportional with the number of cores, that doesn't help either.


SpecInt_RateBase2006:
2s/6c Xeon 5680(3.33) = 355.
4s/6c/ Opteron 8439 SE(2.8 ) = 328.

SpecFP_RateBase2006:
2s/6c Xeon 5680(3.33) = 248.
4s/6c/ Opteron 8439 SE(2.8 ) = 261.
4s/6c/ Opteron 8435 (2.6) = 248.

Java SPECjbb_2005
2s/6c Xeon 5680(3.33) = 928393.
4s/6c/ Opteron 8439 SE(2.8 ) = 1045732.
4s/6c/ Opteron 8435 (2.6) = 926697.

[nettmd]

I can't promise overclockable.

If you can promise me Magnycours will be cheaper, have higher performance AND an overclockable motherboard then you will make me an AMD fan for life.

Wasn't s7e9h3n able to do some nice overclocking om MC back in Septemper??

http://www.xtremesystems.org/forums/...ighlight=cores

[JF-AMD]

I believe I found the answer why Magny Cours was priced so low, less than 1k for the top high end part. The reasons is more than obvious when you look at the performance : it is bound to suck and cost a lot SW wise. Considering that for a lot of SW your license fee is proportional with the number of cores, that doesn't help either.


SpecInt_RateBase2006:
2s/6c Xeon 5680(3.33) = 355.
4s/6c/ Opteron 8439 SE(2.8 ) = 328.

SpecFP_RateBase2006:
2s/6c Xeon 5680(3.33) = 248.
4s/6c/ Opteron 8439 SE(2.8 ) = 261.
4s/6c/ Opteron 8435 (2.6) = 248.

Java SPECjbb_2005
2s/6c Xeon 5680(3.33) = 928393.
4s/6c/ Opteron 8439 SE(2.8 ) = 1045732.
4s/6c/ Opteron 8435 (2.6) = 926697.

Well, I wouldn't count on software licenses. ~80% of the sw is licensed by the socket, by the server or by the named user. We looked into this pretty heavily. The only 2 holdouts are Oracle and VMware Vsphere. But both of those companies are pushing site licenses, so in that case core counts become a non-issue again. There are a few applications that license by thread, so that means you turn HT off. Since Intel folks always like to talk about 50% performance gains, that really means a huge performance hit. If you go by my numbers (10-20% performance uptick) then turning it off is a no-brainer.

Trust me, we've been down the software path and analyzed the numbers, we're fine.

Now, if you are going to bring up cost, I will ask, how many server customers buy those $1669 processors? (spoiler alert: not many) The reality is that most people buy down the stack. You'll see great performance from us at the top of the stack, in many cases better than the numbers you put in above, but where we will really shine is down the stack. THAT is where the volume of the market is. That is where Intel only has quad core choices and the quad core westmeres happen to have only 1066 memory and slower QPI.

It's all about targeting the right part of the market.

[qurious63ss]

I don't want to rain on their parade, let them have their day in the sun. You'll all know soon enough. I am smiling big time today.

When you look at the die size of Instanbul 346mm2 and realize that you have to put two dice together to make magny cores that's a whopping 700mm2 of silicon going up against a mere 250mm2 of Intel silicon. Oh yeah and on top of that you have to pay GF to make those chips for you. If your into making money that would probably wipe that smile right off, but if you could care less about margins then by all means keep on smiling.

[JF-AMD]

There is a lot more to the business than just margins. And I am betting, as a customer myself, that most would rather have savings passed on to the customer vs. paying to a company in the form of higher profits.

Yes, we pay GF to make our chips, but we no longer spend billions to build fabs. There is a net savings for customers.

[Sam_oslo]

Here is my first attempt with X5650, on air:

[qurious63ss]

There is a lot more to the business than just margins. And I am betting, as a customer myself, that most would rather have savings passed on to the customer vs. paying to a company in the form of higher profits.

Yes, we pay GF to make our chips, but we no longer spend billions to build fabs. There is a net savings for customers.

Well that explains why Amd has lost so much money the last several years. They are selling their chips cheap because they want to pass the saving to the customers and not because they can only sell them at those price points.

[anubis]

hey guys, chill pill?

[affiliate13]

Here is my first attempt with X5650, on air:

When you have time, could we see some higher BCLK figures please?
Thanks.

[trn]

Anyone know the differences between the w3680 and the i7-980x? Does the Xeon have the same unlocked multi?

[Movieman]

hey guys, chill pill?

You took the words right out of my mouth.
Same old story, someone posts the thread on one product, the roaches come out of the woodwork to dump on it.
I'm sitting here asking myself what has to happen for this crap to stop?
Do we have to block all the fanboi's from the news,AMD and Intel sections?
Discussion is one thing. Blatant trolling and intentional misinformation is another.

[onex]

hm..
this dual magny-cours's @2992 has managed to bypass 2x980's @3315 on wPrime 1024m..





..

Edit:
sorry, missed u MovieMan..

[qurious63ss]

If your job is to sell Amd chips then you probably will dump on an Intel chip every chance you get. I wonder if he gets extra for posting as much as he does, I could swear he's in every tech forum on the web.

[-Sweeper_]

hm..
this dual magny-cours's @2992 has managed to bypass 2x980's @3315 on wPrime 1024m..
..

Edit:
hh.. sorry, missed u MovieMan..

too bad the fastest magny cours will be clocked at 2.3ghz uh?

[terrace215]

http://it.anandtech.com/printarticle.aspx?i=3769

So what can we conclude so far? Add two cores and a few tweaks to a server CPU architecture which already has the fastest cores on the market and you’ll get very impressive results. Right now, the six-core Xeon wins every comparison with a similar dual CPU configuration. The interesting thing to note is that the margin varies heavily with the type of application.

ERP applications and OLTP databases benefit a lot from the increased L3-cache, hyperthreading and the extra cores. The result is that those applications show absolutely stunning results for Intel: the dual CPU platform is just as fast as AMD best quad CPU configurations. With twice the amount of performance per core there is simply no other option than Intel.

The a similar picture appears for the well scaling native applications such as OLAP or DSS databases. The Xeon 5670 did not slaughter the competition there, but it was still significantly ahead. Be aware though that many native applications will only scale well in certain scenarios. Database size, usage patterns, disk performance and other factors must all be considered. It is not because your application runs on Oracle or SQL Server that it will automatically make good use of the extra cores and threads. A single six-core Xeon will be fast enough in a lot of cases and a second CPU might only add 30% or so.


And...


12 Westmere cores are enough to come very close to the performance of a 24-core Opteron machine. This is does not bode well for the newest octal and twelve-core Opterons (Magny-cours). To be really frank, we think the SAP market is Intel owned until AMD launches the multi-threaded Bulldozer CPU. Most of the SAP server market is not very sensitive to pricing, let alone CPU pricing. SAP projects, which need expensive licenses and many consulting hours are typically in the $100K to $100M range and x86 hardware costs are most of the time only a small percentage of the total project costs. The final blow is the appearance of the Nehalem EX at the end of this month.

Factor in Mangy-Cores' low clockspeeds, and it ain't gonna be pretty.