[XBL] AMD will launch 12-core in March

[Nedjo]

http://www.xbitlabs.com/news/cpu/dis...h_Sources.html

Advanced Micro Devices will launch its twelve-core AMD Opteron “Magny-Cours” central processing units (CPUs) for high-end servers in March, 2010, sources close to the company revealed. AMD plans to unveil a broad family of twelve- and eight-core chips for multi-processor machines with different performance and thermal design power points

[skugpezz]

only if i could afford it

[vengance_01]

these will be just MMC and not native chips. Not a bad idea for a stop gap till bulldozer

[Sam_oslo]

A "broad family of twelve- and eight-core chips", and some of these for desktop? seams like a good move to counter Intel's upcoming 6-cores.

Lets hope these gets into desktop too, because they can do the same to Intel-CPU prices that 4870X2 did to Nvidia GPU-prices.

[informal]

these will be just MMC and not native chips. Not a bad idea for a stop gap till bulldozer

Yeah a 12 and 8 core monolithic MPUs done on 45nm node is real smart... Not.
The MCM via DirectConnect is much much more efficient than the one intel done via FSB in the past. Not only it's efficient but is also cost effective.AMD already bins EE parts,now they will take the best of the best of those and make EE 12 and 8 core ones,much simpler than to test and validate the hypothetical monolithic 12/8 core dice ( uneconomical and taking huge die area with a much more possibilities for defects). With the MagnyC. MCM you will get a 4 channel DDR3 IMC too,the good part of the way AMD does MCM.

[003]

A "broad family of twelve- and eight-core chips", and some of these for desktop? seams like a good move to counter Intel's upcoming 6-cores.

Lets hope these gets into desktop too, because they can do the same to Intel-CPU prices that 4870X2 did to Nvidia GPU-prices.

Except adding cores doesn't really do much after a point. They need to step up IPC, and step down power consumption. Intel CPUs are faster clock for clock and use less power to boot. That is what really hampers AMD CPUs.

[nullface]

Except adding cores doesn't really do much after a point. They need to step up IPC, and step down power consumption. Intel CPUs are faster clock for clock and use less power to boot. That is what really hampers AMD CPUs.

What?!. Even if the use less power to boot (again.. what?!) witch they might do, i dont know, it cant possible be a very significant factor to choose one over the other. "Boot power consumption"

[Sam_oslo]

Except adding cores doesn't really do much after a point. They need to step up IPC, and step down power consumption. Intel CPUs are faster clock for clock and use less power to boot. That is what really hampers AMD CPUs.

Both Intel and AMD quads are overkill for "normal" desktop use, anyways. Therefore they perform almost the same for desktop-applications and game-use.

Those who really need CPU-power, using CPU-hungry applications (like 3D-render, video encoding/decoding, and alike ... ) won't get better performance of 12core 2GHZ than 6core 3GHZ?.

[003]

Both Intel and AMD quads are overkill for "normal" desktop use, anyways. Therefore they perform almost the same for desktop-applications and game-use.

Those who really need CPU-power, using CPU-hungry applications (like 3D-render, video encoding/decoding, and alike ... ) won't get better performance of 12core 2GHZ than 6core 3GHZ?.

I wouldn't be surprised if a 6-core Westmere based Xeon with HT would be enough to beat or come very very close to the performance of a native 12-core opteron, and even beat it in a number of scenarios.

That is currently the case with Nehalem-EX vs current Opterons. In fact Westmere has a slightly higher IPC than Bloomfield so that is another consideration.

That really says something about AMD's performance and efficiency.

[ecidious]

Both Intel and AMD quads are overkill for "normal" desktop use, anyways. Therefore they perform almost the same for desktop-applications and game-use.

Those who really need CPU-power, using CPU-hungry applications (like 3D-render, video encoding/decoding, and alike ... ) won't get better performance of 12core 2GHZ than 6core 3GHZ?.

That depends more on the software than anything; most programs suffer from diminishing returns the more cores you add due to inefficiency with many cores. for example, rendering/distributed computing could be faster if you split those twelve cores into three different VM's and so the programming sees each as a quadcore.

[zalbard]

Server only, so not so interesting I guess.
Only 2.2GHz, though... So not really useful unless you have some very well threaded applications.

[v0dka]

Except adding cores doesn't really do much after a point. They need to step up IPC, and step down power consumption. Intel CPUs are faster clock for clock and use less power to boot. That is what really hampers AMD CPUs.

You did notice that this thread is about server cpu's. Right?

[NickF]

I wouldn't be surprised if a 6-core Westmere based Xeon with HT would be enough to beat or come very very close to the performance of a native 12-core opteron, and even beat it in a number of scenarios.

That is currently the case with Nehalem-EX vs current Opterons. In fact Westmere has a slightly higher IPC than Bloomfield so that is another consideration.

That really says something about AMD's performance and efficiency.

I highly doubt that in server applications a 6 core 12-thread chip could come close to a native 12-core chip. In single-threaded applications, Intel wins. But where it counts in the server world, AMD won't be touched. Especially with new in-house north bridges. Desktop world, Intel is going to continue to dominate for the foreseeable future.

[jcool]

I highly doubt that in server applications a 6 core 12-thread chip could come close to a native 12-core chip. In single-threaded applications, Intel wins. But where it counts in the server world, AMD won't be touched. Especially with new in-house north bridges. Desktop world, Intel is going to continue to dominate for the foreseeable future.

In case you guys are interested in an early comparison between the two, there are some benches of a DP Magny-Cours (so 24 cores) done by s7e9h3n here vs. some benchies of a DP Westmere (12 cores / 24 threads) done by me here

Of course, it's too early to tell, but it seems that clock-for-clock, the Magny-Cours will slightly beat the Westmere in heavily multithreaded apps, while I suspect the Westmere will clock higher due to 32nm process. It also wins in memory-intensive tasks due to the higher bandwidth of the LGA1366 platform and the Tri-Channel DDR3 IMC.

All in all, I'd love to run both for crunching

[informal]

In case you guys are interested in an early comparison between the two, there are some benches of a DP Magny-Cours (so 24 cores) done by s7e9h3n here vs. some benchies of a DP Westmere (12 cores / 24 threads) done by me here

Of course, it's too early to tell, but it seems that clock-for-clock, the Magny-Cours will slightly beat the Westmere in heavily multithreaded apps, while I suspect the Westmere will clock higher due to 32nm process. It also wins in memory-intensive tasks due to the higher bandwidth of the LGA1366 platform and the Tri-Channel DDR3 IMC.

All in all, I'd love to run both for crunching

MC will have 4 ch. DDR3 IMC.

[jcool]

Really? That's even better then! I just hope mobos/chips won't be too expensive (note that expensive is relative with DPs.. Intels aren't exactly cheap either lol)

[accord99]

I highly doubt that in server applications a 6 core 12-thread chip could come close to a native 12-core chip. In single-threaded applications, Intel wins. But where it counts in the server world, AMD won't be touched. Especially with new in-house north bridges. Desktop world, Intel is going to continue to dominate for the foreseeable future.

Why not, 4 core/ 8 thread Nehalem is already easily the match for a 6 core Istanbul at the same clockspeed. And since Magny Cours will be much more limited in clock speed versus 32nm Gulftown combined with Amdahl's Law, it seems very likely that Gulftown will be capable of matching the multi-threaded throughput of Magny Cours.

[ajaidev]

In case you guys are interested in an early comparison between the two, there are some benches of a DP Magny-Cours (so 24 cores) done by s7e9h3n here vs. some benchies of a DP Westmere (12 cores / 24 threads) done by me here

Of course, it's too early to tell, but it seems that clock-for-clock, the Magny-Cours will slightly beat the Westmere in heavily multithreaded apps, while I suspect the Westmere will clock higher due to 32nm process. It also wins in memory-intensive tasks due to the higher bandwidth of the LGA1366 platform and the Tri-Channel DDR3 IMC.

All in all, I'd love to run both for crunching

http://www.hwbot.org/signature.img?i...35&thumb=false

http://img223.imageshack.us/img223/8140/wprime155a.jpg

500Mhz is not a slight difference in my book. Both you people are using ram clocked around 700Mhz, you are using 12GB and he is using 8GB..

If this 12 core opty does sell around $1k mark intel has a problem. Too bad i only have one i9 aka i7 980 "Intel this name sucks BTW" and a desktop variety mobo

[jcool]

Well it's only one app... can't draw a final conclusion from wprime (old V. 1,55 no less).
I would like to see a Fryrender run on Stephen's part, it really is optimized for state-of-the-art hardware.

All I could find was this overclocked Quad Istanbul which also has 24 cores, and is clocked similarly high vs the DP Westmere

[gallag]

what about the old performance per watt metric? If the performance is close between westmere and Magny-Cours It is pretty obvious what one will be sucking more power.

[003]

I highly doubt that in server applications a 6 core 12-thread chip could come close to a native 12-core chip.

Nehalem-EX already does that See the linked sources posted in the thread.

[jcool]

what about the old performance per watt metric? If the performance is close between westmere and Magny-Cours It is pretty obvious what one will be sucking more power.

If I had to guess, I'd say Westmeres are more power-efficient. Due to their 32nm process, they draw less power than a Bloomfield (with identical clocks/voltages) despite having 50% more cores/threads. I can't see the Magny-Cours using less power than current Phenom II's with it having 3x the core count and the same 45nm process.

[ajaidev]

If I had to guess, I'd say Westmeres are more power-efficient. Due to their 32nm process, they draw less power than a Bloomfield (with identical clocks/voltages) despite having 50% more cores/threads. I can't see the Magny-Cours using less power than current Phenom II's with it having 3x the core count and the same 45nm process.

+1

But the 32nm is still in its starting stages, where in Q1 2010 AMD's 45nm will be as good as it could get. Even then i dont think AMD can really do much against Intel because of such a huge core count.

What matters is overall system consumption and blade performance/watt....
If its bad AMD chip will sell cheap cheap cheap, great for us well the ones who have G34 mobo's

[Particle]

Why not, 4 core/ 8 thread Nehalem is already easily the match for a 6 core Istanbul at the same clockspeed. And since Magny Cours will be much more limited in clock speed versus 32nm Gulftown combined with Amdahl's Law, it seems very likely that Gulftown will be capable of matching the multi-threaded throughput of Magny Cours.

Just as a quickie using CB10, I've found my Istanbuls at home to outperform the Nehalems at work. Yay extra cores! Boo tests that suck beyond about 8 cores.

[jcool]

Yeah, boo Cinebench 10. Especially since it doesn't support more than 16 threads. We really need a CB11.

Had to shut off HT to get a proper result