AMD QuadFather matches Kentsfield

[StyM]

source here (http://www.theinquirer.net/default.aspx?article=34918)


The retail boxes are coming in a bit larger size than current packaging and feature an open look at two CPUs, but for the first time, AMD will be launching not one, but three FX CPU packages. All of the CPUs will feature 256KB of L1 and 2MB of L2 cache memory (each core has 64+64KB L1 and 1MB L2 cache), and be placed in three speed grades.

[Charles Wirth]

Read the BS carefully

Match "price and clock speed"...

[FghtinIrshNvrDi]

hmm... good they at least have a temporary answer to the c2d architecture until their own new arch comes out.

Ryan

[K404]

I`m agreeing with Fugger... no performance comparisons. Clock speed matching does not performance euphoria make.

[informal]

Already posted before.
http://www.xtremesystems.org/forums/showpost.php?p=1760744&postcount=140

[STEvil]

the inq, yay!

not.

[theteamaqua]

FX-74 is clocked at 3.0 GHz and finally breaks the 3 GHz clock barrier for desktop and dual-core CPUs in general. However, the price is currently planned at sky-high $1500 !!!!

i aint paying that much money for a CPU that has 2 cores useless to me (in games)

[turtle]

Come on, it's a Saturday...Were you expecting something exciting to break from a more interesting source?


i aint paying that much money for a CPU that has 2 cores useless to me (in games)

understandable.

[specv]

FX-74 is clocked at 3.0 GHz and finally breaks the 3 GHz clock barrier for desktop and dual-core CPUs in general. However, the price is currently planned at sky-high $1500 !!!!

i aint paying that much money for a CPU that has 2 cores useless to me (in games)

$3000 for the fx-74 4x4?

id still take a x6800 that

[turtle]

$1500 for a set of two. They're sold in pairs.

[Vikodemous]

$3000? http://img.photobucket.com/albums/v208/bevelheadgrl/thud.gif

[Charles Wirth]

$3000 for 4x4 cores.

Any leak on clovers MSRP?

[nn_step]

$1500 for a set of two. They're sold in pairs.

what he said

[Vapor]

Yeah, not $3000 unless you build two systems.

FUGGER, "4x4" is just a name, no 8- or 16-core configs available just yet (only 4 cores for now, spread across two CPUs). It's $750 per 3GHz CPU, but you MUST buy them in pairs of two (which sucks, what if one doesn't OC well? Gotta buy another set of two and try to recoup losses on the fleabay market....:slapass: )

And while they may clock past what Kentsfield does at stock, well, we don't care so much about stock performance here :p:

Cloverton's MSRPs were leaked actually.....let me see if I can dig them up.

EDIT: http://www.dailytech.com/article.aspx?newsid=4253

Price-per-core, per-estimated-performance (where 3GHz AMD = 2.66GHz Intel) is also lower on 4x4.

$1500 is a lot for tech that isn't the best and dumps a lot of heat :(

[coldpower27]

$3000 for 4x4 cores.

Any leak on clovers MSRP?

How did you come up with $3000?

AMD would have no chance if this were true, anyway the cost is for a PAIR of Athlon FX-74's is 1500USD. Which makes more sense.

A pair of Athlon FX-70's aka 2 Dual Cores @ 2.6GHZ is 999USD.
A pair of Athlon FX-72's aka 2 Dual Cores @ 2.8GHZ is 1132USD.
A pair of Athlon FX-74's aka 2 Dual Cores @ 3.0GHZ is 1500USD.

Not bad considering the Athlon 64x2 5200+ is 403USD, so the FX-70 represents a 97US price premium for an extra 2x512KB of cache, though they have to be bought in pairs, and are for Socket F only.

This is in relation to the Core 2 Extreme QX6700 for 999USD.

Clovertown MCM Quad Core X5355 @ 2.66GHZ is going to be 1172USD for 1. So a pair is going to be a bit over 2300USD.

[turtle]

Yeah, but one can think optimistically for the future...


$1500 is a lot for tech that isn't the best and dumps a lot of heat :(

True, but $1500 sounds fairly enticing for 2x X4 2.7ghz Altair FX's...Even if it's a year away. :rolleyes:

Screw it...You're right. Yorkfield XE will probably be cheaper, cooler, and probably a gigahertz faster.

I hope K8L is one hell of an architecture, for AMD's sake.

[Fred_Pohl]

source here (http://www.theinquirer.net/default.aspx?article=34918)

I wonder if a $1500 pair of FX74s might match the performance of a $1000 Kentsfield. Probably not but it might be close as long as you don't overclock.

[coldpower27]

Yeah, but one can think optimistically for the future...



True, but $1500 sounds fairly enticing for 2x X4 2.7ghz Altair FX's...Even if it's a year away. :rolleyes:

Screw it...You're right. Yorkfield XE will probably be cheaper, cooler, and probably a gigahertz faster.

I hope K8L is one hell of an architecture, for AMD's sake.

The only reason this is so "cheap" is because it is 2 Dual Cores facing 1 Quad Core.

Once we have an Altair/Barcelona FX which is going to be 2 Quad Core processor for this 4x4 platform, there won't be anything stopping AMD from charging 2000USD or more for a PAIR of them.

Since it will be an 8 Core platform 2x Altair/Bareclona vs a 4 Core platform 1x Yorkfield.

[Vapor]

I wonder if a $1500 pair of FX74s might match the performance of a $1000 Kentsfield. Probably not but it might be close as long as you don't overclock.
It'll be pretty close with performance....at stock.

The next issue is motherboards. How much will they cost? How good will they ACTUALLY be? Any decent OCing options?


Clovertown MCM Quad Core X5355 @ 2.66GHZ is going to be 1172USD for 1. So a pair is going to be a bit over 2300USD.Yeah, a pair of those will also be 8 cores with about the same heat dump as two FX-72s. Same situation with the motherboards though, but heck, it's 8 cores.

If Smithfield was a quick-job, so is this (all they did was take an existing product line, add a few SKUs and new names and a new marketing name). Further down the road, like Smithfield's dual-die design (seen in Kentsfield), there will be advantages though....once they make it more flexible.

Being able to buy one moderately-priced Altair, then another a bit later would be a tremendous bonus, IMO.

This may be a novel step in the right direction, but like Smithfield, it's just not going to compete in the short-term and will need some major finessing to work in the long-term.

And I agree....no reason to rule out $2000 pairs later :(

[coldpower27]

It'll be pretty close with performance....at stock.

The next issue is motherboards. How much will they cost? How good will they ACTUALLY be? Any decent OCing options?

Yeah, a pair of those will also be 8 cores with about the same heat dump as two FX-72s. Same situation with the motherboards though, but heck, it's 8 cores.

If Smithfield was a quick-job, so is this (all they did was take an existing product line, add a few SKUs and new names and a new marketing name). Further down the road, like Smithfield's dual-die design (seen in Kentsfield), there will be advantages though....once they make it more flexible.

Being able to buy one moderately-priced Altair, then another a bit later would be a tremendous bonus, IMO.

This may be a novel step in the right direction, but like Smithfield, it's just not going to compete in the short-term and will need some major finessing to work in the long-term.

And I agree....no reason to rule out $2000 pairs later :(


By the time we have Altair for the Athlon 64x4 line in a 1P system, there will be decent Core 2 Quads Kentsfield, and soon likely Yorkfields as well. Not to mention intel since they are using MCM will have the yield advantage over AMD's near 300mm2 die size "native/monolithic" Quad Cores.

I dunno, to me 1500USD for a pair of Athlon FX-74's is hardly a deal compared to the Quad Core 999USD Kentsfield. Not to mention each processor on the 2P FX platform will need it's own set of memory and the possibility of requiring registered stuff.

Note: oops after reading your post again Vapor i apologize i am not responding correctly. Read the stuff below for something more accurate.

Yes, I agree this is basically taking to Santa Ana Opteron 22xx Series cores and validating them for operation on desktop platform. Intel is also somewhat guilty of this too however, with their Xeon 3000 line.

We will have to see if their will be moderately priced Pairs of Athlon FX's down the line. However I doubt it.

[Charles Wirth]

Ah, I was confused by a previous slide, my bad. Somehow I missed the lesser was in pairs and the top end sold in pair for less.

[afireinside]

Not to mention each processor on the 2P FX platform will need it's own set of memory and the possibility of requiring registered stuff.


At least you won't have to deal with cores starving of bandwidth and shared cache killing performance with 2P FX...

[nn_step]

Not to mention each processor on the 2P FX platform will need it's own set of memory and the possibility of requiring registered stuff.
Yes each Processor will need its own memory, and that will be the cause for all AMD processors for a long time. However the need for Registered memory is incorrect.
At the same time 4x4 has lower latency then Intel's MCM
Simply because the cores can directly talk to each other. Intel's design on the other hand means that one processor die talks to the northbridge and then that northbridge talks to the other processor die. And if you didn't know it makes :banana::banana::banana::banana:ty performance, cache conflicts and on a poorly coded OS write errors

[theteamaqua]

oh so its 2 x FX-74 for $1500, but its still a lot of money for 1 comp .... i can built 2 E6700 rig OC to 3.8 , 3.9Ghz ... and it might not be slower.

[sierra_bound]

Clovertown MCM Quad Core X5355 @ 2.66GHZ is going to be 1172USD for 1. So a pair is going to be a bit over 2300USD.
Clovertown is meant for server boards. You can't really put in the same category as CPU's designed to go in boards that can be overclocked.

Most businesses running servers can easily afford $2300.

[Charles Wirth]

Im on a Mac Pro?? call me crazy but I could use Clovertowns yesterday/last month :D

[sierra_bound]

A couple of people here have shown screens with dual Clovertowns in Supermicro boards. The results were pretty impressive for systems that weren't even overclocked. The rendering power was awe-inspiring.

[coldpower27]

At least you won't have to deal with cores starving of bandwidth and shared cache killing performance with 2P FX...

The problem with this is that there is currently no indication that Core architecture or AMD K8 that the cores on both respectively lines are bandwidth starved.

The shared cache in Intel's case helps performance as it still represents to 2MB of LV2 per core on Kentsfield.

[coldpower27]

Clovertown is meant for server boards. You can't really put in the same category as CPU's designed to go in boards that can be overclocked.

Most businesses running servers can easily afford $2300.

I was only responding in the fact that someone posted and inquired on what the price of Clovertown was, and I answered that. I didn't imply that Clovertown was meant to go head to head with the 4x4 intiative.

Yes, and I agree the cost structure ofr a business would be different compared to the consumer.

[nn_step]

I was only responding in the fact that someone posted and inquired on what the price of Clovertown was, and I answered that. I didn't imply that Clovertown was meant to go head to head with the 4x4 intiative.

Yes, and I agree the cost structure ofr a business would be different compared to the consumer.
Actually according to Intel, it is just supposed to be a stop gap measure until Yorkfield arrives :rolleyes: The first thing that pops into my mind is Pentium D and the fact that Windows XP can NOT issue more than 4 Threads to anymore than 2 Cores at a given time (unless M$ released a patch for that and forgot to tell me and forgot to update their website)

[coldpower27]

Yes each Processor will need its own memory, and that will be the cause for all AMD processors for a long time. However the need for Registered memory is incorrect.
At the same time 4x4 has lower latency then Intel's MCM
Simply because the cores can directly talk to each other. Intel's design on the other hand means that one processor die talks to the northbridge and then that northbridge talks to the other processor die. And if you didn't know it makes :banana::banana::banana::banana:ty performance, cache conflicts and on a poorly coded OS write errors

If you had read my post you would have seen I said the possibility of requiring registered memory. I didn't say it was a 100% sure thing, as currently there are rumors flying about for both situations.

How so? the 2P 4x4 intiative still needs to communicate through the HyperTransport link, since they are in different Sockets, I do not call this directly, while Kentsfield will communicate through the FSB.

Nice try though, even arguing from the point that AMD has a "more elegant" form of processor communication, what matters will be overall performance of the processor as well as cost.

Having shared cache actually increases peformance as each set of 2 cores have a unifed cache pool to draw from. I really wonder if a problem will arise regarding shared cache, as Intel who has the bulk majority of processors or the OS vendor will quickly address such an issue if it actually exisits, not to mention AMD is also moving to shared cache designs as well.

[coldpower27]

Actually according to Intel, it is just supposed to be a stop gap measure until Yorkfield arrives :rolleyes: The first thing that pops into my mind is Pentium D and the fact that Windows XP can NOT issue more than 4 Threads to anymore than 2 Cores at a given time (unless M$ released a patch for that and forgot to tell me and forgot to update their website)

A link? Clovertown is a first generation Quad Core for the 2P segment. It is quite a viable measure as for multithreaded programs which can use all the cores available there are substantial gains. It will have no competition so even if it was a "stop gap" as you so eloquently put it, it is the only option for 2P Quad Core at its time of introduction and hence doesn't need to be the best, once AMD provides competition with Barcelona Intel can release their update to this with Harpertown.

Yorkfield is actually suppose to be the replacement for Kentsfield later on in 2007, and the server 2P version of that is called Harpertown.

What about the Pentium D, so Intel quickly made a Dual Core solution to combat the X2, it's better then having NO solution whatsoever. At least it had some merit as at the time AMD was charging high prices for it's Dual Cores. Intel provided lower price Dual Cores for the mianstream effectively. The Socket F Athlon FX line will be as expensive as Kentsfield XE for the cheapest package, as well as limiting you to Socket F processors only.

Microsoft Windows XP Pro is allowed to utilize up to 2 Sockets worth of processors. So you can run a Clovertown based system on XP, with all of the 8 Cores being shown on the task manager. I seriously doubt that XP Pro has such a limitation if Clovertown was shown on Xtremesystems.

[Darkenreaper57]

4x4 looks very nice, but it is way too expensive for the average enthusiast, imho.

Unless K8L is cheaper, it looks like I'll buy an intel system next.

[kemo]

That is certainly a bad news Intel is preparing affordable quad cores that should perform better than 4X4 especially when overclocked then AMD release 4X4 at high prices and still no body knows how the boards will cost that is disappointing only few fan boys will boy the FX-74
________
Ford Falcon (North America) (http://www.ford-wiki.com/wiki/Ford_Falcon_(North_America))

[Lightman]

...
Not to mention each processor on the 2P FX platform will need it's own set of memory and the possibility of requiring registered stuff.
...

Untrue! Everyone seems to forget that Opterons can use memory from other CPU using cHT. Look at old S940 boards where you had one memory bank routed to CPU0 and no memory bank to CPU1. Anyone with Opteron mobo can take out dimms from mobo and left only 1 CPU with mem and it will work! Of course when each CPU have own memory pool then you get full BANDWIDTH from 2 mem controllers.

[zakelwe]

So 4x4 is going to have to use motherboards from "stability" biased motherboard makers such as Tyan and Iwill? I seem to recall from running dual AMD alongtime ago for F@H that these were poor for overclocking on FSB and multipliers .. voltage adjustments were a joke.

Core2 duo or quadro overclocks like a :banana::banana::banana::banana::banana: at the drop of a hat in the best enthusiast motherboards, I don't care how much FSB they are starved or theoretical limitations, if you can hit 3600Mhz on a Kentsfield then a $1500 FX-74 setup is going to be stomped if it can only hit 3.3Ghz due to motherboard limitations. Clock for clock core is 20% faster than AMD at least so AMD need 4.3Ghz out of quad FX if Kentsfield easily hits 3600.

I know Kentsfield gets ripped for being two dual cores pasted together but at least you have it all in one package.

Say I suggest some hypothetical air cooled XS style Kentsfield system

Kentsfield quad core
Asus motherboard
2GB memory from Corsair / OCX / Geil / Team ? / etc
1x Tuniq tower hsf

All that is pretty mainstream I suggest. What do the AMD quadfathers suggest as direct competition ? What air cooling can you fit on the board for instance as a first question?

Regards
Andy

[Lightman]

...
Core2 duo or quadro overclocks like a :banana::banana::banana::banana::banana: at the drop of a hat in the best enthusiast motherboards, I don't care how much FSB they are starved or theoretical limitations, if you can hit 3600Mhz on a Kentsfield then a $1500 FX-74 setup is going to be stomped if it can only hit 3.3Ghz due to motherboard limitations. Clock for clock core is 20% faster than AMD at least so AMD need 4.3Ghz out of quad FX if Kentsfield easily hits 3600.
....

Say I suggest some hypothetical air cooled XS style Kentsfield system

Kentsfield quad core
Asus motherboard
2GB memory from Corsair / OCX / Geil / Team ? / etc
1x Tuniq tower hsf

All that is pretty mainstream I suggest. What do the AMD quadfathers suggest as direct competition ? What air cooling can you fit on the board for instance as a first question?

Regards
Andy


Good point! 4x4 will be wasted if mobo makers won't support it :mad:.

If manufactures will stand and deliver good 4x4 mobos with HT3.0 and all tweaks you can find in mainstream ASUS or DFI boards, then it could be more like investment for future. Today buy cheapest FX-70 and OC it to extreme, after 9 months go to shop and buy K8L quad core, put to your system and be happy! Otherwise all this is simply to expensive for most enthusiasts!

As for cooling I didn't hear about performance cooling solutions for S1207 so only option will be BOX cooler, at least on beginning. AMD will need to put really good cooler into this 4x4 sets of FX's!

[AACDIRECT]

Good point! 4x4 will be wasted if mobo makers won't support it :mad:.

If manufactures will stand and deliver good 4x4 mobos with HT3.0 and all tweaks you can find in mainstream ASUS or DFI boards, then it could be more like investment for future. Today buy cheapest FX-70 and OC it to extreme, after 9 months go to shop and buy K8L quad core, put to your system and be happy! Otherwise all this is simply to expensive for most enthusiasts!

As for cooling I didn't hear about performance cooling solutions for S1207 so only option will be BOX cooler, at least on beginning. AMD will need to put really good cooler into this 4x4 sets of FX's!

I would not be suprised to see a board from Asus. They are one of AMD's major partners.

For cooling no one has info on the mount format for Socket F; it could be that the same HSF that fits AM2 (which is the same as 939/940/754) fit it. With how well AMD looks to backward compatiblity again I would not be suprised. (At least that is my hope.)

[zakelwe]

I would not be suprised to see a board from Asus. They are one of AMD's major partners.

For cooling no one has info on the mount format for Socket F; it could be that the same HSF that fits AM2 (which is the same as 939/940/754) fit it. With how well AMD looks to backward compatiblity again I would not be suprised. (At least that is my hope.)

Even if it the same mount will you be able to physcially fit TWO of the big heatpipe heatsinks that have taken air cooling to a new level ? With Kentsfield you can also easily fit single phase change or water, with 4x4 you obviously have to double up on cooling.

Regards

Andy

[Vapor]

Yeah, as much grief as Intel's MCM gets, AMD's doing it too, just the two dies are a few inches apart :(

Even if there were good boards, K8s don't clock very well. Even without AMD's inferior IPC....Kentsfield has already clocked, on air, faster than ANY AMD dual-core has ever clocked, regardless of cooling. This is just a patch-up job to not fall too far behind.

And as I've said, 4x4, like MCM, will have benefits later, but for now it's just a desperation move and these prices are semi-ludicrous for it.

[AACDIRECT]

Even if it the same mount will you be able to physcially fit TWO of the big heatpipe heatsinks that have taken air cooling to a new level ? With Kentsfield you can also easily fit single phase change or water, with 4x4 you obviously have to double up on cooling.

Regards

Andy

That will be the trick. Guess it will depend on the size of the MB (ATX Vs E-ATX) , and layout. Even with water-cooling there will be the need to cool two physical cpus. It is going to be more complicated than a single cpu for sure. For those with Phase it will be the biggest pain.

One positive I can think of is that maybe you can OC each CPU seperately. I know at one time AMD was working on allowing CPUs of different speeds to be run in DP. If so that could be kind of interesting.

[FghtinIrshNvrDi]

I kinda hate to see AMD wasting marketing $$$ and development time/money on something that won't quite keep up with a single chip Intel is putting out. I'd rather see those guys go and work on K8L.

Ryan

[AACDIRECT]

I kinda hate to see AMD wasting marketing $$$ and development time/money on something that won't quite keep up with a single chip Intel is putting out. I'd rather see those guys go and work on K8L.

Ryan

One thing I would say is safe to assume is 4 X 4 will be compatible with K8L. When is Intel roadmapped to have a consumer 8 core setup? AMD might be behind now but IF K8L is good AND will drop into a 4 X 4 MB then that could be something.

[arisythila]

I believe these FX's are just like FX's before them(unlocked multiplier), I dont think its going to be ahrd to break 3 GHZ on air on the lower line procs.. I would also be surprized if you cant buy them seperately. I remeber AMD saying this was possible... Worse case, I guess we could just find out who wants one then and there, and then go halves...


~Mike

[Dublin_Gunner]

If you had read my post you would have seen I said the possibility of requiring registered memory. I didn't say it was a 100% sure thing, as currently there are rumors flying about for both situations.

How so? the 2P 4x4 intiative still needs to communicate through the HyperTransport link, since they are in different Sockets, I do not call this directly, while Kentsfield will communicate through the FSB.

Nice try though, even arguing from the point that AMD has a "more elegant" form of processor communication, what matters will be overall performance of the processor as well as cost.

Having shared cache actually increases peformance as each set of 2 cores have a unifed cache pool to draw from. I really wonder if a problem will arise regarding shared cache, as Intel who has the bulk majority of processors or the OS vendor will quickly address such an issue if it actually exisits, not to mention AMD is also moving to shared cache designs as well.

in defense of nn_step (not that he needs defending), its quite apparent you have no clue what you're talking about, and have no clue about the Opteron system architecture.

HTT is NOT like an FSB, as you appear to have that view.

On a server type AMD board (which is what 4x4 is), there are more than 1 HTT link.
1 link for CPU to external i/o etc communications, and at least 1 more link for CPU to CPU communications.

Each CPU will probably have 3 HTT links, just like opterons.

Also shared cache can be as detrimantal to performance as it can be beneficial. It can lead to a hell of a lot of cache thrashing, which destroys performance.

i.e 1 Core has cache filled with data it needs, 2nd corfe checks cache for data it needs, doesnt find it, wastes cycles clearing cache to be filled with data from main memory, continues with task.
But, Core 1 checks back to cache looking for the data Core 2 has just cleared, and must waste more cycles clearing the cache again, to use it for its own task, then and so on.

Both cores end up constantly wiping / filiing cache, and searching from main memory, defeating the purpose of having cache therer in the first place.

Of course, this would only generally happen if 1 core used up the majority of cache space for its own task, and the other core was doing something completely un-related

[metro.cl]

One thing I would say is safe to assume is 4 X 4 will be compatible with K8L. When is Intel roadmapped to have a consumer 8 core setup? AMD might be behind now but IF K8L is good AND will drop into a 4 X 4 MB then that could be something.

But then if k8l is a good chip and can fit 4x4 you have the high $$$$ problem.


long time ago talking to lev i was really mas about 4x4 and i remember telling him about overclock and saying to him, "well try putting ln2. dry ice or phasa change to it" its gonna suck for hardcore overclockers

[Dublin_Gunner]

I agreee. A server type board with 2 CPU's is a pain to cool. Either you have 2x Large hiogh performance air coolers, or you have to try use 2 phase nuits, or getting extra pupms for your water cooling.

Its gonna get messy inside that case.


I doubt 4x4 will even launch properly. Its such a retarded concept.


I mean, how is it that new anyway? Multi CPU systems have been around years. And I'd much rather have 1xquad core, than 2x dual core cpu's.

[nn_step]

I agreee. A server type board with 2 CPU's is a pain to cool. Either you have 2x Large hiogh performance air coolers, or you have to try use 2 phase nuits, or getting extra pupms for your water cooling.

Its gonna get messy inside that case.


I doubt 4x4 will even launch properly. Its such a retarded concept.


I mean, how is it that new anyway? Multi CPU systems have been around years. And I'd much rather have 1xquad core, than 2x dual core cpu's.
Very true, although i wouldn't really call MCM'd set of Conroes a quad core, more along the lines of a set of Dual cores working on the same socket.
Lacking the ability to communicate to each other directly. AMD to has mentioned the possible use of MCM, however what they are planning on doing is connecting 2 dies via the HTT connection and giving each of them a single channel. However after looking at the benefits and drawbacks of this, I HOPE they go with a trinity chip option. Which is the usage of a Third Extremely specialized chip, which will contain a HTT controller and a Dual channel Memory controller. Which then acts as the shared northbridge of their earlier designs but being external will add a few ns of latency but nothing as horrible as intel's design.

[metro.cl]

from all I know so far 4x4 won't be aimed at overclockers, but at PC enthusiasts, and most of them don't really overclock. There are those ppl that care for performance, but do not wish to care for the details. Like I might want a fast PC without having LN2 agregates and stuff to think about.

people that are interested in performance but not in technology will love a stable platform, I really don;t think the majority of those ppl will enjoy random reboots as they use the PC cause they didn't gave enough volts on the GPU or CPU or chipset, deal with mods, and so on....

I thing 4x4 might get more attention from users that require powerfull workstations at a good price point. Still noone knows how 4x4 will scale, and how it will perform compared to Kentsfiled

Making two chip quad core is easier than native quad core. But dual dualcore setup is even easier to make IMO, as it doesn't require new design

some people here can't realize overclockers make a very small percent of PC sales, and both intel and amd are concernet about making money a lot more than how their chips overclock and perform. Right now AMD would kill intel if they can offer a tiny core that has half the performance of a kentsfileld, but costs AMD 1$ to make it. Uber performance and overclocking are too smal - about 10% of the sales. AND it is about how much you can make and sell in most of the cases.

for all of that you'll need extremely multi threaded software, and so far that isnt the case. Software companies will code for the mainstream and mainstream is still single core so that is why so few are investing aditional resources in multithreaded apps.

[Theli]

One thing I would say is safe to assume is 4 X 4 will be compatible with K8L. When is Intel roadmapped to have a consumer 8 core setup?
Isn't that the Clovertown (http://www.xtremesystems.org/forums/showthread.php?t=111251)?
I may have missed something, though.

[DilTech]

Im on a Mac Pro?? call me crazy but I could use Clovertowns yesterday/last month :D

Ok FUGGER, I'll do it.

You're crazy! :lol: :ROTF: :rofl: :nuts:

[Haltech]

So which motherboard choices does someone have? Is it going to require ECC ram or what?

[nn_step]

So which motherboard choices does someone have? Is it going to require ECC ram or what?
it is going to use unbuffered ram, ecc isn't required but it is an option

[zabomb4163]

shouldnt the title of the thread be fixed?

[[cTx]Philosophy]

This all came from the INQ fellas, Ive learned not to take what they post anywhere to much into interest of mine..

While the quadfather does seem like an overpriced attempt to steal some of the intel fanbase back to AMD, this is just something they threw together to make it hold over to K8L, where there hopefully gonna make the performance gap a little closer..

This is intels race to lose here..

[Dublin_Gunner]

@ LOE

If someone wanted a workstation setup, they would buy a workstation board and 2 Opterons.

4x4 is quite clearly aimed at gaming enthusiasts, which in the most part DO enjoy overclocking and tinkering.

I still think its a stupid idea, dressing up something thats been around for years as 'new'

[awdrifter]

I wonder if this means we can use the 22XX series Opterons on the 4x4 mobo, if we could, that would be pretty kick ass.

[awdrifter]

You can get two 1.8ghz 2210 Opteron for $270 each (assuming they work), so that's only $540, much cheaper than getting the FX. I always thought the point of 4x4 is to having a cheaper 2 socket baord that's overclocking friendly. People overclock their Opteron 165 to 2.6-2.8ghz, so it would be pretty kick ass to have two of them.

[Dublin_Gunner]

I think its quite apparent that these 'FX' chips are merely re-branded Opterons with the Multi unlocked.

There is no other reason for them to use the Socket F form factor.

No FX chips have ever been designed for multiple CPU setups. SO how do you get around this?

Get your Opty's, unlock the multis (I asssume as they're FX, they'll have unlocked multi's?), and re-brand them FX chips. That way you already have stock of these CPU's with the xtra HTT links needed for multiple CPU set-ups, but virtually no dev cost.

Its even been stated that the only people releasing 4x4 boards will be the high end server board manufacturers, highlighting further that 4x4 is nothing more than a re-branded workstation setup. Re-branded board & CPU.

Its rubbish.


@ LOE, if it wasnt aimed at overclockers & enthusiasts, why release only 'FX' branded chips for the set-up? Everyone knows the main reason for an FX cpu is the unlocked multi, of which the only use is overclocking.

[Dublin_Gunner]

Problem with AM2 is the lack of extra HTT buses.

No provision for multi CPU systems so it simply would not be technically possible.

[The Ghost]

Problem with AM2 is the lack of extra HTT buses.
do we know this for sure ?

[Dublin_Gunner]

Yes.

AM2 just like s939 is a single CPU system. 1 HTT link. the CPU's have only 1 HTT llink on die, and the boards only support 1 HTT link.

Multiple HTT links have been the preserve of server / opteron based systems (although NOT 939 variants of Opterons)

[Dublin_Gunner]

Why would they do that?

If you wanted [AMD]multi processor systems previously, you would have to buy an Opteron system.

Allowing multiple CPU's on AM2 or 939 would eat into their server profit margins.

You dont see Intel dishing out multi CPU boards for desktop systems, thats reserved for server applications.

[The Ghost]

Yes.

AM2 just like s939 is a single CPU system. 1 HTT link. the CPU's have only 1 HTT llink on die, and the boards only support 1 HTT link.
and just how do we know this for sure ? all amd has said is that they are disabled on non server cpu's

remember what the socket 940 was before it became the desktop cpu , so there is enough pins

[Dublin_Gunner]

If you knew anything about K8 based systems, you would know this for sure.

s940 Opterons had multiple HTT links, for server applications.

s754, 939 & AM2 have the extra links either disabled or they're not even present.

Why would you release a nice cheap desktop chip so someone could place it in a server box?? That would be plain stupid.

Only Opteron 2xx series and upwards have multiple HTT links

[The Ghost]

If you knew anything about K8 based systems, you would know this for sure.

s940 Opterons had multiple HTT links, for server applications.

s754, 939 & AM2 have the extra links either disabled or they're not even present.

Why would you release a nice cheap desktop chip so someone could place it in a server box?? That would be plain stupid.

Only Opteron 2xx series and upwards have multiple HTT links
i do know about K8 systems

on socket 754 , 939 and 940 the extra links are disabled , or they are suppose to be , who has tested them to make sure that they are disabled being that there is not a motherboard that would support them if they was not disabled , who is going to test them to make sure that they are actually disabled ??

a desktop cpu does not fit in a server dual cpu motherboard not even a 1xx or a 1xxx cpu will fit in a dual cpu motherboard , , so back to it again , who can be sure that the extra htt has been disabled , other then amd saying so ?

[Dublin_Gunner]

If thats the case, does it even matter?

Considering you can use the CPU's in those systems anyway...:stick:

[The Ghost]

If thats the case, does it even matter?

Considering you can use the CPU's in those systems anyway...:stick:
actually it does matter if amd was to use am2 socket for the 4x4 motherboard set up , this would allow people that have am2 cpu's could use the dual socket motherboard , so for amd to stop this , they are using the socket f motherboards which are made for running two cpu's or more

so who are we to say for sure that am2 cpu's do not have the extra htt links ?

[Dublin_Gunner]

Opterons have the extra HTT links, desktop CPU's do not.

You may contact AMD to confirm this if you like.

Also, you'll notice no desktop CPU's will work in server boards, so even if they are not disabled, you wont be able to find out.

Either way, no desktop boards come equiped with more than 1 HT bus, so its a moot point.

As I've stated in earlier posts, its quite apparent that all 4x4 is, is a re-branded server set-up, nothing more.

If this was indeed possible with AM2 CPU's, this would obviously be a much favourable method to implement it for AMD, as they would already have potential buyers with half of the hardware required.

The fact that their using Socket F boards & CPU's, simply highlights the fact that not only do AM2 cpus NOT have the extra HTT links needed, but the development costs of AM2 boards with extra HTT links would be too costly, and they would have to get board partners to develop them.

Conclusion -

AM2 = 1 HTT
939 = 1 HTT
754 = 1 HTT
s940 = Multiple HTT's
s F = Multiple HTT's

[nn_step]

Opterons have the extra HTT links, desktop CPU's do not.

You may contact AMD to confirm this if you like.

Also, you'll notice no desktop CPU's will work in server boards, so even if they are not disabled, you wont be able to find out.

Either way, no desktop boards come equiped with more than 1 HT bus, so its a moot point.

As I've stated in earlier posts, its quite apparent that all 4x4 is, is a re-branded server set-up, nothing more.

If this was indeed possible with AM2 CPU's, this would obviously be a much favourable method to implement it for AMD, as they would already have potential buyers with half of the hardware required.

The fact that their using Socket F boards & CPU's, simply highlights the fact that not only do AM2 cpus NOT have the extra HTT links needed, but the development costs of AM2 boards with extra HTT links would be too costly, and they would have to get board partners to develop them.

Conclusion -

AM2 = 1 HTT
939 = 1 HTT
754 = 1 HTT
s940 = Multiple HTT's
s F = Multiple HTT's

Correction

AM2, s939, s754 = 1 Coherent HTT and 2 non-Coherent HTT
s940 = 1-3 Coherent HTT links with the total number of links not to exceed 3
sF = 1-4 Coherent HTT links with the total number of links not to exceed 4

Plus HTT 3.0 Specification states that all HTT 3.0 compliant 16bit links can split into 2 8bit HTT links. Since all AMD's HTT links are 16bit.. thus
socket AM2+ (aka AM2 w/ HTT 3.0) has 2 coherent HTT links which is more than enough for a 2P system :stick:

[The Ghost]

actually this going to use a server type socket shows that there is a possibility that the extra htt links have not been cut off , there has been no one to prove otherwise , because there is not a motherboard for am2 , 939 , or 754 motherboard , all we have is amd's word , by amd switching to socket F might be the proof that the htt links are not disabled

the proof is in the pudding , because if the htt links was not disabled , amd could use the dual socket am2 motherboard

now just think it thru fgor a while

[Dublin_Gunner]

all that using socket F proves is that you're basically buying a server platform, with re-branded opterons.

nn_step, that fact that the desktop HTT links are non-coherant, they may as well not be present at all, as they cannot be utilised.


I have not seen any HTT 3.0 compliant hardware on the market yet, so my points remain valid, as we were discussing AM2

[The Ghost]

all that using socket F proves is that you're basically buying a server platform, with re-branded opterons.
rebranded opterons ? where have you seen a opteron that is 3ghz that is going to be rebranded for the 4x4 setup ?


I have not seen any HTT 3.0 compliant hardware on the market yet, so my points remain valid, as we were discussing AM2
am2 is going to have htt 3 on it

[awdrifter]

The fastest Opteron right now is at 2.8ghz, so I don't think getting 0.2ghz out of a hand picked chip is that hard. I agree that this 4x4 stuff is just rebranded Opteron.

[uOpt]

So which motherboard choices does someone have? Is it going to require ECC ram or what?

All AMD4 CPUs support but do not require ECC RAM.

I know that some people are still living in a dream world, but it is certain by now that 4x4 will require registered RAM.

They will not invent a new type of socket by making a platform that uses the same physical socket as socket F but uses unregistered RAM. That's just insane. If they did that you couldn't use 4x4 FXes in socket F boards and vice versa. AMD can't do that, the vendors will rip their heads off. It's not going to happen, better start shopping for registered D9 modules now.

%%

I think two 3.0 GHz AMD64 dual-cores are reasonably attractive at $1500, but I'm not going to buy CPU pairs. What if I step on one CPU when putting it in? Buy a new pair?

%%

The sad part is that the only advantage this will have over Kentsfields will be in highly optimized multithreaded applications, which split up single tasks (as opposed to giving each core a different task). That rules out all games. It's also not the case when you run multiple different applications.

They say they match by price/MHz and Core2 is 20% better than AMD64 (single-thread) at the same clockspeed, even with the worst applications. So you would have to make good 20% speed in interprocessor communications (see above) or overclocking potential.

%%

And Tyan and Supermicro will offer overclocking boards with all the voltage options we want. Right. That's going to happen. I think only Asus might be a player here, and then the (non-) competition situation kills the project in pricing.

[zakelwe]

It'll be interesting to see the persons face who is trying to plug all the power leads into the 4x4 when using 2 x G80 as well :D

Right so 24pin for there, and an 8 pin to there and the 4pin to there and another 6 pin to there and an 8 pin to there and another 6 pin to there an yet another 8 pin to there. Phew!

The Intel version is no better either !

Is there a power supply that has that many 12v cables ? :confused: :eek: :D


Regards

Andy

[nn_step]

It'll be interesting to see the persons face who is trying to plug all the power leads into the 4x4 when using 2 x G80 as well :D

Right so 24pin for there, and an 8 pin to there and the 4pin to there and another 6 pin to there and an 8 pin to there and another 6 pin to there an yet another 8 pin to there. Phew!

The Intel version is no better either !

Is there a power supply that has that many 12v cables ? :confused: :eek: :D


Regards

Andy
Nope but something tells me we are not going to need that many anyways

[STEvil]

I've been saying how dumb the multiple standards on PSU's is forever.. it sucks :(

[uOpt]

I've been saying how dumb the multiple standards on PSU's is forever.. it sucks :(

The only thing that is stupid is having the 4+4 EPS12V connector on PSUs like OCZ and the like do. PSUs from Seasonic and Zippy have separate 4-pin and 8-pin EPS12V.

The 4+4 EPS12V is actually triple stupid because the "wrong" half of it will fit into the painboard 4-pin plug - just 90 degrees turned (sending power into ground pins).