Question Beyond Rome: AMD's EPYC and Radeon to Power World's Fastest Exascale Supercomputer

[PaulAlcorn]

AMD's EPYC CPUs and Radeon GPUs have been selected to power the world's fastest exascale supercomputer.

Beyond Rome: AMD's EPYC and Radeon to Power World's Fastest Exascale Supercomputer : Read more

 

[Deleted member 14196]

Nice!

 

[akamateau]

nVidia seems to be blocked from the two top Supercomputers.

 

[animalosity]

But will it run Crysis?

 

[meriax1029]

SkyNet goes live 2021. Thanks AMD!

 

[digitalgriffin]

That explains where the 7nm thread ripper dies went...

What's curious here is the choice of GPU's. Instinct has excellent FP throughput. It's no match for NVIDIA's GV100 on AI metrics where matrix math is required.

It means they are not going to be using it for AI type math. So no learning algorithms. Most likely cell based algorithms that take more time to process each node the further you break it down.


��

Tesla V100 (SXM2)	Tesla V100 (PCIe)
Double Precision	7.5 TFLOPS	7 TFLOPS
Tensor Performance (Deep Learning)	120 TFLOPS	112 TFLOPS
GPU	GV100

MI25[edit]
The MI25 is a Vega based card, utilizing HBM2 memory. The MI25 performance is expected to be 12.3 TFLOPS using FP32 numbers. In contrast to the MI6 and MI8, the MI25 is able to increase performance when using lower precision numbers, and accordingly is expected to reach 24.6 TFLOPS when using FP16 numbers. The MI25 is rated at <300W TDP with passive cooling. The MI25 also provides 768 GFLOPS peak double precision (FP64) at 1/16th rate.[5]

 

[prtskg]

That explains where the 7nm thread ripper dies went...

What's curious here is the choice of GPU's. Instinct has excellent FP throughput. It's no match for NVIDIA's GV100 on AI metrics where matrix math is required.

It means they are not going to be using it for AI type math. So no learning algorithms. Most likely cell based algorithms that take more time to process each node the further you break it down.


��

Tesla V100 (SXM2)	Tesla V100 (PCIe)
Double Precision	7.5 TFLOPS	7 TFLOPS
Tensor Performance (Deep Learning)	120 TFLOPS	112 TFLOPS
GPU	GV100

MI25[edit]
The MI25 is a Vega based card, utilizing HBM2 memory. The MI25 performance is expected to be 12.3 TFLOPS using FP32 numbers. In contrast to the MI6 and MI8, the MI25 is able to increase performance when using lower precision numbers, and accordingly is expected to reach 24.6 TFLOPS when using FP16 numbers. The MI25 is rated at <300W TDP with passive cooling. The MI25 also provides 768 GFLOPS peak double precision (FP64) at 1/16th rate.[5]

Wouldn't MI50 and 60 would provide better comaprison?
https://www.tomshardware.com/news/amd-radeon-instinct-mi60-mi50-7nm-gpus,38031.html
MI60 has 7.4TFLOPS of FP64 performance and very efficient (perhaps one of the most efficient right now) for HPC.

 

[digitalgriffin]

Wouldn't MI50 and 60 would provide better comaprison?
https://www.tomshardware.com/news/amd-radeon-instinct-mi60-mi50-7nm-gpus,38031.html
MI60 has 7.4TFLOPS of FP64 performance and very efficient (perhaps one of the most efficient right now) for HPC.

You're quite right. Good catch. It's still lacking though on the AI side.

 

[bit_user]

SkyNet goes live 2021. Thanks AMD!

Why single out AMD? As the article cited, an Intel-based machine was already announced - also slated to come online in 2021.

And what's so special about 1.5 exaflops? Why is that the magic number for building Skynet?

 

[bit_user]

What's curious here is the choice of GPU's. Instinct has excellent FP throughput. It's no match for NVIDIA's GV100 on AI metrics where matrix math is required. It means they are not going to be using it for AI type math. So no learning algorithms.

This won't be a current-gen GPU, so it will probably also have something like tensor cores. In Anandtech's coverage, they specifically cited AI as a target workload:

Accordingly, the lab is expecting the supercomputer to be used for a wide range of projects across numerous disciplines, including not only traditional modeling and simulation tasks, but also more data-driven techniques for artificial intelligence and data analytics.

https://www.anandtech.com/show/1430...ntier-supercomputer-cray-and-amd-1-5-exaflops

And it's worth noting that Nvidia's Tensor cores aren't general-purpose matrix math processors - they're too low precision for most uses of matrix math, really only accelerating matrix multiplications of the sort used in deep learning.

��

Tesla V100 (SXM2)	Tesla V100 (PCIe)
Double Precision	7.5 TFLOPS	7 TFLOPS
Tensor Performance (Deep Learning)	120 TFLOPS	112 TFLOPS
GPU	GV100

MI25[edit]
The MI25 is a Vega based card, utilizing HBM2 memory. The MI25 performance is expected to be 12.3 TFLOPS using FP32 numbers. In contrast to the MI6 and MI8, the MI25 is able to increase performance when using lower precision numbers, and accordingly is expected to reach 24.6 TFLOPS when using FP16 numbers. The MI25 is rated at <300W TDP with passive cooling. The MI25 also provides 768 GFLOPS peak double precision (FP64) at 1/16th rate.[5]

Why are you comparing with MI25? MI60 is based on the latest 7nm Vega, matches the V100 on fp64 and beats it on memory bandwidth.

Still, it's not really a great comparison, since both companies will be at least one generation beyond those chips, by 2021.

 

[prtskg]

This won't be a current-gen GPU, so it will probably also have something like tensor cores. In Anandtech's coverage, they specifically cited AI as a target workload:


https://www.anandtech.com/show/1430...ntier-supercomputer-cray-and-amd-1-5-exaflops


Still, it's not really a great comparison, since both companies will be at least one generation beyond those chips, by 2021.

True that

 

[TCA_ChinChin]

Why single out AMD? As the article cited, an Intel-based machine was already announced - also slated to come online in 2021.

I think its just because normally, it would be mostly just Intel based machines, but its news simply because AMD has been able to get such high performance machines on the road. Also it'll be the worlds most powerful, which is something worth noting.

 

[mavroxur]

Toms should double check their math. It won't be faster than the top 160 systems combined. It's 2.5x faster than #2 (Fukagu) and is about as fast as the #2-#8 systems combined. Not #2-#160

 

[USAFRet]

Toms should double check their math. It won't be faster than the top 160 systems combined. It's 2.5x faster than #2 (Fukagu) and is about as fast as the #2-#8 systems combined. Not #2-#160

3 years later.....

Please don't dredge up ancient threads.