[XBL] AMD will launch 12-core in March

[FischOderAal]

only if i could afford it

That's always the problem with all this great stuff

[flippin_waffles]

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.

Are you kidding? Under the workloads these things will be used for, a 12 core Magny Cours will destroy 6 cores with hyperthreading. For this kind of workload, nehalem's rendition of hyperthreading is no different than P4's was for crying out loud. If the software is properly optimized, hyperthreading shouldn't help the bloody nehalem one single bit. Do you feel that the world class enginneers working at AMD have absolutely no idea what they're doing, and should have done things your way all along?

[Florinmocanu]

12 cores vs 6 cores is definitely not an even match, a 12 core magny cours will be a monster of a CPU. I sure would want one for my 3dsmax renderings...

/drool

[gOJDO]

[Manicdan]

whats that a picture of?

[accord99]

Are you kidding? Under the workloads these things will be used for, a 12 core Magny Cours will destroy 6 cores with hyperthreading.

Only if the Magny Cours runs at the same clock speed as Gulftown, but I suspect Gulftown will be capable of running at much higher clockspeeds.

For this kind of workload, nehalem's rendition of hyperthreading is no different than P4's was for crying out loud. If the software is properly optimized, hyperthreading shouldn't help the bloody nehalem one single bit.

Actually, for sever workloads Hyperthreading is generally very helpful. And just because software is optimized doesn't mean it won't hit bottlenecks outside of the CPU or fully utilize all the resources of a Nehalem core.

Do you feel that the world class enginneers working at AMD have absolutely no idea what they're doing, and should have done things your way all along?

They're making lemonade out of lemons.

12 cores vs 6 cores is definitely not an even match,

All things being equal.

[flippin_waffles]

Only if the Magny Cours runs at the same clock speed as Gulftown, but I suspect Gulftown will be capable of running at much higher clockspeeds.


Actually, for sever workloads Hyperthreading is generally very helpful. And just because software is optimized doesn't mean it won't hit bottlenecks outside of the CPU or fully utilize all the resources of a Nehalem core.


They're making lemonade out of lemons.


All things being equal.

You can try and nip that bud all you like, but the fact of the matter is 6 cores and hyperthreading isn't going to scale anywhere near as well as 12 cores in code that can utilize 12 cores. And given that AMD's 2 chip design scales almost linearly, those 6 cores from intel are going to need a big clock advantage. Interesting that it's now intel that needs to get the clock rate up to match AMD's performance.

[jcool]

Wow, some serious fanboy alert. I'm outta here...

[skugpezz]

Wow, some serious fanboy alert. I'm outta here...

lol

[N19h7m4r3]

whats that a picture of?

AMD Quadcore on the left and Intel Quad on the right. Before i7 ofcourse.

[Chumbucket843]

ever heard of nehalem ex? 8 nehalem cores and 24MB L3 with a ring bus which is better for communication between cores and latency. current designs are not good at scaling with more than 8 cores. hyperthreading is a way to reduce some of these problems. smt is not only used by intel but by a lot of non x86 cpu's. obviously these chips target hpc and servers so its definitely useful. if you think ht is the same in nehalem as it is in p4 you are ignoring fact.

[accord99]

You can try and nip that bud all you like, but the fact of the matter is 6 cores and hyperthreading isn't going to scale anywhere near as well as 12 cores in code that can utilize 12 cores.

But another fact is that a single Gulftown core's base performance will be much higher than a single Magny Cours core. So, it'll end up depending on, as they usually do, relative clock speeds

And given that AMD's 2 chip design scales almost linearly, those 6 cores from intel are going to need a big clock advantage. Interesting that it's now intel that needs to get the clock rate up to match AMD's performance.

Not that large I suspect. And then there's always Beckton coming in from above.

[flippin_waffles]

...I'm outta here...

Ignorance is bliss huh!

[003]

Are you kidding? Under the workloads these things will be used for, a 12 core Magny Cours will destroy 6 cores with hyperthreading. For this kind of workload, nehalem's rendition of hyperthreading is no different than P4's was for crying out loud. If the software is properly optimized, hyperthreading shouldn't help the bloody nehalem one single bit. Do you feel that the world class enginneers working at AMD have absolutely no idea what they're doing, and should have done things your way all along?

Uh no actually I'm not kidding. You can try and argue this all you want but there was already a very big review conducted by Anandtech of a 12-core Nehalem-EX system vs a 24-core Opteron system and Nehalem-EX came very close.

See this post and this review by Anandtech of Nehalem-EX vs Shanghai.

[Manicdan]

anad doing intel vs amd articles feels like a bigger fanbio thread than most i see on here.

[flippin_waffles]

Uh no actually I'm not kidding. You can try and argue this all you want but there was already a very big review conducted by Anandtech of a 12-core Nehalem-EX system vs a 24-core Opteron system and Nehalem-EX came very close.

See this post and this review by Anandtech of Nehalem-EX vs Shanghai.

Well that's all well and good, but like I said before, how about comparing software that is written specifically and optimized with parallelism in mind. It's pretty hard to compare since there really isn't any sort of benchmark out there yet that is truly optimised for multicore. But it looks like that is just around the corner. So for poorly coded and last gen software, intel has an advantage. Not something i'd want to brag about though since the future software is being written with multicore in mind, so hyperthreading's value is going to go into the toilet where it belongs... Scalability is where it's at.

[initialised]

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

That view is so this decade...

[god_43]

i like the tdps on those chips!

[Chumbucket843]

Well that's all well and good, but like I said before, how about comparing software that is written specifically and optimized with parallelism in mind. It's pretty hard to compare since there really isn't any sort of benchmark out there yet that is truly optimised for multicore. But it looks like that is just around the corner. So for poorly coded and last gen software, intel has an advantage. Not something i'd want to brag about though since the future software is being written with multicore in mind, so hyperthreading's value is going to go into the toilet where it belongs... Scalability is where it's at.

that doesnt even make sense. HT increases multithreaded performance by reducing latency and bubbles and only increases die size by 5%. nehalem scales very well so all of these posts from you on HT/nehalem that simply are not true need to stop. if you think chip architects at intel are so incompetent then why dont you teach them how to do their jobs? i just find it strange that you can distort data from real world benchmarks to end up in amd's favor when it isnt.

there are benchmarks and programs optimized for multicore already just not many core. current software infrastructure is not ready for that.

[Chrono Detector]

Would love to see a desktop version of this. 12 cores = pure win

[flippin_waffles]

that doesnt even make sense. HT increases multithreaded performance by reducing latency and bubbles and only increases die size by 5%. nehalem scales very well so all of these posts from you on HT/nehalem that simply are not true need to stop. if you think chip architects at intel are so incompetent then why dont you teach them how to do their jobs? i just find it strange that you can distort data from real world benchmarks to end up in amd's favor when it isnt.

there are benchmarks and programs optimized for multicore already just not many core. current software infrastructure is not ready for that.

I'm not trying to distort anything. My initial point was meant to be that magny cours should have no problem outperforming 6 cores plus 6 threads in highly parallel work loads, ie in the environment they are designed for.
Surely you agree that hardware like the 4800's in the Chinese supercomputer announced recently has software taylor made to use the hardware at peak efficiency.

So I don't see how an extra thread adding a maximum of %40 performance under ideal conditions, is not going to have a disadvantage against a real core adding 2x the resources. in highly parallel workloads

[FlanK3r]

flippin_waffles: right. Some guys here have eyes only for Intel...Did u seen Stephans benchmarks with Magny? In wprime was Nehalem at 5 GHz sooo bad than Magny at 3 GHz . I hope to future, AMD kick ass Intels CPU again (P4 vs K8).

[gOJDO]

whats that a picture of?

That is a picture of Henri Richard, a marketing clown employed by AMD showing a months old CPU from Intel(Kentsfield) which is a MCM of 2 native dualcores on the right and a non-working CPU(Barcelona) of AMD, released months later which was a native quadcore on the left. After so much hype, lies and uncertainty, Barcelona was so late, so slow and buggy.

That clown, as well as many others, who were forced to invest their reputation(if they had any at all) to support the AMD failures were BS-ing about the MCM approach. With Magny Cours, being a native MCM AMD finally is confirming that Intel was going to the right direction.

flippin_waffles: right. Some guys here have eyes only for Intel...

And some have only for AMD.

In wprime was Nehalem at 5 GHz sooo bad than Magny at 3 GHz .

So what? A Core2 Duo E8200 at 2.66GHz outperforms Magny Cours @5GHz in Super Pi 1M!

I hope to future, AMD kick ass Intels CPU again (P4 vs K8).

Hope dies last. AMD needs a new and much better architecture, as well as new and much better production process if they want to outperform Intel's x86 CPUs.

[FlanK3r]

Do u think, wprime is old benchmark as superpi? Optimized wprime for multitasking is more better (here is i7 with HTT>i5>=Phenom II X4>Core Quad etc...)

[003]

a post full of fail

Rather than trolling as you usually do, are you capable of pointing out specific remarks in the post that you believe are false or untrue and explain why you believe so?