A keynote from Raja wouldn't be worth a byline on any other website, yet here it's top billing deposing a review of the best consumer x86_64 CPU on the planet. Dr Intel - mission accomplished. Sorry, Ian. It's Ian. Yep. definitely Ian.
It's a lame comment, but I don't really see what rules it's breaking. I don't think comments should be removed or people banned because you disagree with them.
Rather than moderation, I think what you really want is up/down-voting buttons.
you forget that intel hired the same guys that made AMD great in the first place, but it takes 4-5 years for their work to come to fruition.
for the time being intel has no response whatsoever, but when jim keller's and raja's work is finally revealed i expect a major comeback. i don't see intel going anywhere else besides chiplets too, especially since jim was very closely involved in amd's zen development.
You're out of your mind if you think Jim Keller was the key to Zen. Just like he was for Apple A4, right? Oh right, he wasn't. That guy who stayed at Apple through to the A12 was. Keller's money trail for the past 8 years has been that of a multi-million dollar consultant.
thats because any other website just goes after clicks and right now the only thing that gets clicks is either praising amd or hating intel. he helped bring amd up to where it should have been years ago and now hes going to bring intel where they should have been years ago, not really threatening amd so much as nvidia.
the 3950x is a great cpu but its just a sku, and one that most people wont buy. its impact will be short lived. in a couple years more performance will be available for cheaper as it always has been.
but this could be the start of something new, many skus. i think its deserving of top billing. the timing might be a strange but intel controls that so what can ian do?
As the beginning of the article said, this is an annual thing for them - at the start of the Supercomputing conference. You might want to dial down that cynicism, just a little. It'll rot your brain.
The story did mention using CXL for cache coherent connections among the CPUs and GPUs within the node, which I presume is also on PCIE5 based on their previous discussions. However, several articles mention use of Slingshort interconnects, which I presume is for between system nodes. For example, this article:
So the empire is getting ready to strike back. Nvidia, AMD, and Intel have been converging for years, and now the showdown is coming. AMD was first to recognize the way the winds were blowing when it bought ATI. It hoped for an early, quick victory with its APU paradigm, but neither software nor market were ready. Intel and Nvidia both drew next at the same time, Intel broadening to massively parallel architectures with Larrabee, Nvidia to centralized ones with ARM and Tegra. Both failed, while AMD whiffed with Bulldozer, but now everybody's back for round 2. Intel has been honing its parallel game, and now seems to have learned the importance of the software stack from Nvidia. Meanwhile AMD has regained its hardware mojo and even scored a win with its packaging innovations. Nvidia still has the software lead, but without a real CPU story it has to rely on the big serial era coming to a close fast enough for this not to matter. AMD has the hardware now but lack of vision and execution on the software side is a major weakness. Finally, Intel remains a hardware force, and its history of strong compilers bodes well for its software chances, but it has yet to prove itself in parallel computing. Let the game begin!
Nvidia killed the APU, or at least complete inability to compete with Intel+Nvidia killed all talk of "heterogenous computing" from AMD. AMD's Picasso and Raven Ridge certainly have made the ALU a strong choice (especially for notebooks).
Raja is going to have far more enemies trying to undermine anything that isn't x86 from inside Intel than he ever imagined at AMD. While they may prefer an internal solution to partnering with nvidia, I still don't expect Intel culture to roll over and accept a non-x86 architecture as important.
Was NVDA even an option? Could NVDA deliver a gpu with PCIE5 and CXL and Intel 7nm ( == tsmc 5nm), and the software support for OneAPI for a 2021 target?
So the Intel message is "Wait, just wait short hopefully 2 more years, do not buy anything from AMD, or you'll deeply regret till the end of your life. We promise, we swear you 3x increase, then 150x".
Are you the Argonne national lab? If you aren't, this is very much not the next thing on the roadmap. AMD sells supercomputers with components that are very far off too, that's just how HPC works because of the massive lead times.
Heard many evaluating AMD 64core EPIC now. I'd also try it with the PIC codes. Less external interconnections and more on the chip itself and 3-4 times larger L3 caches may give us serious speed boost versus 12core Intel Xeon chips. Basically the only test which show Intel huge 3-4x speed advantage is Ian's own AVX512 3D particle movement test the source of which is not open. PIC code though show just 15% improvement with ACX512
Could you explain this a bit more, about SIMT looking like AVX? Are they saying they are borrowing from the avx encodings/syntax to implement gpu matrix operations? I recall someone writing that there is encoding space available for avx1024 ... did they perhaps grab that encoding and provide broad support for all the current avx512 operations, extending somehow to gpu?
The slide on the Rambo Cache shows 4096 x 4096 DPFP Matrices. Is this meant to be a hard limit in their API? I'm assuming they are handling the tiling.
I doubt that that is a hard limit on the software side as that is straight forward to break down there.
Rather I suspect that the 4096 x 4096 is hitting the cache size limit. Each matrix would be 128 MB in size and there is a need for at least three (two inputs and a result). The real question is just how large those Rambo chips are in terms of capacity.
Right now the implication would be 512 MB, though that could be spread throughout the entire package, not per Rambo die. If that's per die, that'd be ~25 billion transistors just in SRAM logic and wouldn't include any sort of controller. Might be possible for a large monolithic chip at 7 nm. They also have the option of stacking SRAM on top of a controller/router which would reduce the foot print. Stacking would permit Intel to leverage 10 nm or even 14 nm lines for the SRAM as 7 nm fab space would have to be at an ultra premium in 2021.
On the SIMD looking like AVX ... Is he saying the intrinsics look like AVX? Does he mean substituting vectors for the registers? I guess I should just go look at oneAPI. 06:53PM EST - *SIMD looks like AVX, sorry
I like how everyone keeps mentioning "7nm" like its somehow related to everything intel does. Its not.
Put this in perspective, even though AMD beat intel to 7nm, it took them how long to do it considering all the 14nm products Intel was able to put out doing it? That is crazy long life for a product.
In reality, its pretty sad that it took AMD this long to even get ahead given all the chances that had. If intel says it takes them 1 year to 10nm, 2 years to get 7nm up and running..and they had 14nm going for SIX years..that says a lot about how much
put it into perspective AMD kicked intels ass big time with the 64bit chips but went the wrong way in design, at no time in its History has intel who has very deep pockets let up on its unfair competition shenanigans so AMD were left out to dry, they then went and bought ATI which was bleeding, they had GloFoun that was bleeding and a huge infrastructure that was sucking them dry, intel came back with core2 and it was a big deal hell even i bought one, sadly it died after 14 Months (first and last intel chip), AMD were virtually bankrupt after all this yet still they came back with zen @ 14nm, then zen+ at 12nm after 12 months then zen2@7nm 12 months later, currently they are sampling 5nm zen 3 and have a design for 3nm in the works for 2021 I hope intel does come back but i also hope they face major pain first and AMD gets the cash riches it needs to compete properly against Nvidia on the high level, so far 5700xt looks good but they need a 2080ti and beyond beater as well to disrupt the market as much and give all us consumers a better deal, just remember the real winners here are us, we can now buy a 12core CPU at reasonable prices who would have thought that 3 years ago?
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
54 Comments
Back to Article
lasserith - Sunday, November 17, 2019 - link
Who is dreamier? Raja or Ian?Dayman - Sunday, November 17, 2019 - link
Hmm, why not both?ianmills - Sunday, November 17, 2019 - link
LOL at Koduri with his white shirt and flaming white hair. Great contrast to Jensen Huang man in black getupDark vs Light! Who will win?
tipoo - Sunday, November 17, 2019 - link
Obviously #SuBaeJorgp2 - Sunday, November 17, 2019 - link
Don't you mean Sapphire Rapids CPU?del42sa - Monday, November 18, 2019 - link
yes :-) it´s a CPU not GPUsvan1971 - Sunday, November 17, 2019 - link
Intel is announcing anything they can to get to the top of the page.CityBlue - Sunday, November 17, 2019 - link
A keynote from Raja wouldn't be worth a byline on any other website, yet here it's top billing deposing a review of the best consumer x86_64 CPU on the planet. Dr Intel - mission accomplished. Sorry, Ian. It's Ian. Yep. definitely Ian.Cellar Door - Sunday, November 17, 2019 - link
It's low quality comments like this - why even a basic moderator is needed for the comments section.mode_13h - Monday, November 18, 2019 - link
It's a lame comment, but I don't really see what rules it's breaking. I don't think comments should be removed or people banned because you disagree with them.Rather than moderation, I think what you really want is up/down-voting buttons.
Death666Angel - Monday, November 18, 2019 - link
Just give me an "ignore" option. Some disadvantages of vote buttons that can be abused.Railander - Monday, November 18, 2019 - link
you guys are out of your minds.you forget that intel hired the same guys that made AMD great in the first place, but it takes 4-5 years for their work to come to fruition.
for the time being intel has no response whatsoever, but when jim keller's and raja's work is finally revealed i expect a major comeback. i don't see intel going anywhere else besides chiplets too, especially since jim was very closely involved in amd's zen development.
mdriftmeyer - Tuesday, November 19, 2019 - link
You're out of your mind if you think Jim Keller was the key to Zen. Just like he was for Apple A4, right? Oh right, he wasn't. That guy who stayed at Apple through to the A12 was. Keller's money trail for the past 8 years has been that of a multi-million dollar consultant.FreckledTrout - Monday, November 18, 2019 - link
You are talking nonsense.bobhumplick - Thursday, November 21, 2019 - link
thats because any other website just goes after clicks and right now the only thing that gets clicks is either praising amd or hating intel. he helped bring amd up to where it should have been years ago and now hes going to bring intel where they should have been years ago, not really threatening amd so much as nvidia.the 3950x is a great cpu but its just a sku, and one that most people wont buy. its impact will be short lived. in a couple years more performance will be available for cheaper as it always has been.
but this could be the start of something new, many skus. i think its deserving of top billing. the timing might be a strange but intel controls that so what can ian do?
Qasar - Friday, November 22, 2019 - link
" the timing might be a strange but intel controls that so what can ian do? "huh ??
mode_13h - Monday, November 18, 2019 - link
As the beginning of the article said, this is an annual thing for them - at the start of the Supercomputing conference. You might want to dial down that cynicism, just a little. It'll rot your brain.billtohara - Sunday, November 17, 2019 - link
Any comment on how many nodes are in Aurora? And what the node to node interconnect is? They killed omnipath. So ethernet, slingshot or ...?mode_13h - Monday, November 18, 2019 - link
My bet is Ethernet."Never bet against Ethernet."
JayNor - Monday, November 18, 2019 - link
I've read in more than one article that it is SlingshotZoolook13 - Wednesday, November 20, 2019 - link
CXL, it's mentioned in the story.JayNor - Saturday, December 21, 2019 - link
The story did mention using CXL for cache coherent connections among the CPUs and GPUs within the node, which I presume is also on PCIE5 based on their previous discussions. However, several articles mention use of Slingshort interconnects, which I presume is for between system nodes. For example, this article:https://www.hpcwire.com/2019/03/18/its-official-au...
mode_13h - Monday, November 18, 2019 - link
> That's a wrap! Time for Q&A.So... I guess there weren't any good questions? Or maybe just no good answers?
Ian Cutress - Monday, November 18, 2019 - link
Stay tuned. Will be a separate article.mode_13h - Monday, November 18, 2019 - link
Thank you for the coverage!ABR - Monday, November 18, 2019 - link
So the empire is getting ready to strike back. Nvidia, AMD, and Intel have been converging for years, and now the showdown is coming. AMD was first to recognize the way the winds were blowing when it bought ATI. It hoped for an early, quick victory with its APU paradigm, but neither software nor market were ready. Intel and Nvidia both drew next at the same time, Intel broadening to massively parallel architectures with Larrabee, Nvidia to centralized ones with ARM and Tegra. Both failed, while AMD whiffed with Bulldozer, but now everybody's back for round 2. Intel has been honing its parallel game, and now seems to have learned the importance of the software stack from Nvidia. Meanwhile AMD has regained its hardware mojo and even scored a win with its packaging innovations. Nvidia still has the software lead, but without a real CPU story it has to rely on the big serial era coming to a close fast enough for this not to matter. AMD has the hardware now but lack of vision and execution on the software side is a major weakness. Finally, Intel remains a hardware force, and its history of strong compilers bodes well for its software chances, but it has yet to prove itself in parallel computing. Let the game begin!abufrejoval - Tuesday, November 19, 2019 - link
You summed it up in a rather rare mixture of eloquence and succintness: Bravo!lobz - Monday, November 18, 2019 - link
This on 7nm in 2021? I'm calling utter bs here.HollyDOL - Monday, November 18, 2019 - link
Well, they managed to wake up and deliver Core 2 after being kicked long enough, so it's not completely unimaginable.Otoh I want to see mass delivery and enough benchmarks to believe the claim :-)
mr_tawan - Monday, November 18, 2019 - link
So AMD killed the APU and now Raja is trying to bring it back at Intel ?wumpus - Monday, November 18, 2019 - link
Nvidia killed the APU, or at least complete inability to compete with Intel+Nvidia killed all talk of "heterogenous computing" from AMD. AMD's Picasso and Raven Ridge certainly have made the ALU a strong choice (especially for notebooks).Raja is going to have far more enemies trying to undermine anything that isn't x86 from inside Intel than he ever imagined at AMD. While they may prefer an internal solution to partnering with nvidia, I still don't expect Intel culture to roll over and accept a non-x86 architecture as important.
JayNor - Saturday, December 21, 2019 - link
Was NVDA even an option? Could NVDA deliver a gpu with PCIE5 and CXL and Intel 7nm ( == tsmc 5nm), and the software support for OneAPI for a 2021 target?watersb - Monday, November 18, 2019 - link
Thanks for covering this. HPCdel42sa - Monday, November 18, 2019 - link
hope they catch the date with 7nm :-)SanX - Monday, November 18, 2019 - link
So the Intel message is "Wait, just wait short hopefully 2 more years, do not buy anything from AMD, or you'll deeply regret till the end of your life. We promise, we swear you 3x increase, then 150x".Sure. By then either donkey die or shah die
ArcadeEngineer - Monday, November 18, 2019 - link
Are you the Argonne national lab? If you aren't, this is very much not the next thing on the roadmap. AMD sells supercomputers with components that are very far off too, that's just how HPC works because of the massive lead times.SanX - Monday, November 18, 2019 - link
Heard many evaluating AMD 64core EPIC now. I'd also try it with the PIC codes. Less external interconnections and more on the chip itself and 3-4 times larger L3 caches may give us serious speed boost versus 12core Intel Xeon chips. Basically the only test which show Intel huge 3-4x speed advantage is Ian's own AVX512 3D particle movement test the source of which is not open. PIC code though show just 15% improvement with ACX512JayNor - Saturday, December 21, 2019 - link
Raja showed a slide with 2.5x speedup of FastWalsh over GPU by using SIMD+SIMT combined GPU and SIMD.https://www.anandtech.com/show/15123/raja-koduri-a...
SanX - Monday, November 18, 2019 - link
"We promise you 3x increase. Do not believe us? Then even more, we promise you 150x! Now do you believe at least in 3x?"Irata - Tuesday, November 19, 2019 - link
Shouldn't Intel really be more careful with big numbers after the initial Optane promises ?JayNor - Monday, November 18, 2019 - link
Thanks very much for the live blog.Could you explain this a bit more, about SIMT looking like AVX? Are they saying they are borrowing from the avx encodings/syntax to implement gpu matrix operations? I recall someone writing that there is encoding space available for avx1024 ... did they perhaps grab that encoding and provide broad support for all the current avx512 operations, extending somehow to gpu?
06:51PM EST - SIMT looks like AVX
JayNor - Monday, November 18, 2019 - link
The slide on the Rambo Cache shows 4096 x 4096 DPFP Matrices. Is this meant to be a hard limit in their API? I'm assuming they are handling the tiling.Kevin G - Monday, November 18, 2019 - link
I doubt that that is a hard limit on the software side as that is straight forward to break down there.Rather I suspect that the 4096 x 4096 is hitting the cache size limit. Each matrix would be 128 MB in size and there is a need for at least three (two inputs and a result). The real question is just how large those Rambo chips are in terms of capacity.
Right now the implication would be 512 MB, though that could be spread throughout the entire package, not per Rambo die. If that's per die, that'd be ~25 billion transistors just in SRAM logic and wouldn't include any sort of controller. Might be possible for a large monolithic chip at 7 nm. They also have the option of stacking SRAM on top of a controller/router which would reduce the foot print. Stacking would permit Intel to leverage 10 nm or even 14 nm lines for the SRAM as 7 nm fab space would have to be at an ultra premium in 2021.
peevee - Friday, November 22, 2019 - link
384MB of SRAM with multiple (1000s) parallel accesses sounds rather implausible on a single chip.Maybe it is the maximum size of 1 output matrix with 4096x1 and 1x4096 inputs. :)
JayNor - Monday, November 18, 2019 - link
What does the "06:54PM EST - 40x increase in DPFP per EU" line mean? 40x vs what?JayNor - Monday, November 18, 2019 - link
On the SIMD looking like AVX ... Is he saying the intrinsics look like AVX? Does he mean substituting vectors for the registers? I guess I should just go look at oneAPI.06:53PM EST - *SIMD looks like AVX, sorry
https://software.intel.com/sites/landingpage/Intri...
imaheadcase - Monday, November 18, 2019 - link
I like how everyone keeps mentioning "7nm" like its somehow related to everything intel does. Its not.Put this in perspective, even though AMD beat intel to 7nm, it took them how long to do it considering all the 14nm products Intel was able to put out doing it? That is crazy long life for a product.
In reality, its pretty sad that it took AMD this long to even get ahead given all the chances that had. If intel says it takes them 1 year to 10nm, 2 years to get 7nm up and running..and they had 14nm going for SIX years..that says a lot about how much
alufan - Monday, November 18, 2019 - link
put it into perspective AMD kicked intels ass big time with the 64bit chips but went the wrong way in design, at no time in its History has intel who has very deep pockets let up on its unfair competition shenanigans so AMD were left out to dry, they then went and bought ATI which was bleeding, they had GloFoun that was bleeding and a huge infrastructure that was sucking them dry, intel came back with core2 and it was a big deal hell even i bought one, sadly it died after 14 Months (first and last intel chip), AMD were virtually bankrupt after all this yet still they came back with zen @ 14nm, then zen+ at 12nm after 12 months then zen2@7nm 12 months later, currently they are sampling 5nm zen 3 and have a design for 3nm in the works for 2021 I hope intel does come back but i also hope they face major pain first and AMD gets the cash riches it needs to compete properly against Nvidia on the high level, so far 5700xt looks good but they need a 2080ti and beyond beater as well to disrupt the market as much and give all us consumers a better deal, just remember the real winners here are us, we can now buy a 12core CPU at reasonable prices who would have thought that 3 years ago?mryamaguchi - Monday, November 18, 2019 - link
Full stops please. Your reply is a pain to readJayNor - Tuesday, November 19, 2019 - link
"currently they are sampling 5nm zen 3 and have a design for 3nm in the works for 2021"zen3 is N7+ ... just adds a few euv steps.
abufrejoval - Tuesday, November 19, 2019 - link
Too bad you were busy while SA got this one first: https://semiaccurate.com/2019/11/18/centaurs-new-c...It's a great new addition to the race!
peevee - Friday, November 22, 2019 - link
"100Bs of companies at every level"BS it is.
peevee - Friday, November 22, 2019 - link
"Heterogenity math in Intel CPUs, 150x in 6 years"They compare to 2010, which is 9 years ago just in case your calculator is broken.
Saatparvaz - Saturday, December 7, 2019 - link
ساعت پرواز قرن مجری انواع تور های مسافرتیhttps://www.saatparvaz.com/