News Nvidia's Arm-Powered Grace CPU Debuts, Claims 10X Performance Over x86 Servers

rluker5

Distinguished
Jun 23, 2014
605
367
19,260
I read " When tightly coupled with NVIDIA GPUs, a Grace CPU-based system will deliver 10x faster performance than today’s state-of-the-art NVIDIA DGX-based systems, which run on x86 CPUs. " which would likely be gpu based and communication performance improvements. Not quite the same as future arm is 10x current x86.
 
  • Like
Reactions: CerianK and Krotow

Jim90

Distinguished
I read " When tightly coupled with NVIDIA GPUs, a Grace CPU-based system will deliver 10x faster performance than today’s state-of-the-art NVIDIA DGX-based systems, which run on x86 CPUs. " which would likely be gpu based and communication performance improvements. Not quite the same as future arm is 10x current x86.

Exactly. Beware of headlines and marketing. "10x current x86" is about as bullcrap a headline as you can get.
 
  • Like
Reactions: CerianK and Krotow

Houndsteeth

Distinguished
Jul 14, 2006
514
3
19,015
Early 2023.

AMD and Intel (as well as others) have some time to revise their roadmaps to take this new player into account.

I think Intel has been way too complacent sitting on their laurels.

If you think only AMD was offering competition, now we can see both Apple and NVidia taking shots.

For anything that doesn't require outright x86/AMD64 compatibility, these new processors offer a very competitive option.

And (at least) the Apple Silicon has proven to be competitive even taking x86/AMD64 compatibility into account, if they would ever consider licensing or selling processing units outside of their own machines.
 

spongiemaster

Admirable
Dec 12, 2019
2,273
1,277
7,560
I read " When tightly coupled with NVIDIA GPUs, a Grace CPU-based system will deliver 10x faster performance than today’s state-of-the-art NVIDIA DGX-based systems, which run on x86 CPUs. " which would likely be gpu based and communication performance improvements. Not quite the same as future arm is 10x current x86.
The actual statement is:

"Nvidia introduced its Arm-based Grace CPU architecture that it says delivers 10X more performance than today's fastest servers in AI and HPC workloads "

That last part which is no where in your statement is the important part. They're not claiming 10x across the board. The current Epyc CPU's in DGX boxes have no AI specific acceleration. It's not really that hard to believe that two years from now Nvidia could release a purpose built customized CPU that beats today's general purpose CPU's by 10x in specific workloads.
 
  • Like
Reactions: VforV and Makaveli

VforV

Respectable
BANNED
Oct 9, 2019
578
287
2,270
The actual statement is:

"Nvidia introduced its Arm-based Grace CPU architecture that it says delivers 10X more performance than today's fastest servers in AI and HPC workloads "

That last part which is no where in your statement is the important part. They're not claiming 10x across the board. The current Epyc CPU's in DGX boxes have no AI specific acceleration. It's not really that hard to believe that two years from now Nvidia could release a purpose built customized CPU that beats today's general purpose CPU's by 10x in specific workloads.
Exactly, specific AI and HPC workloads. But some people don't know to read apparently...
 

Drazen

Distinguished
Dec 29, 2015
28
8
18,535
Yeah, NVidia and marketing crap.
Do you remember Tegra1 days and how NVidia claimed Tegra1 is faster then Intel Core CPUs?
Of course it will be 1000000 times faster if NVidia runs code written in C++ and x86 Python. But do not tell anyone this unimportant detail :p