facebook rss twitter

AMD launches the 11.5TFLOPS (FP64) Instinct MI100 accelerator

by Mark Tyson on 17 November 2020, 10:11

Tags: AMD (NYSE:AMD)

Quick Link: HEXUS.net/qaepo4

Add to My Vault: x

While PC gamers and enthusiasts are on tenterhooks waiting for the first RDNA 2 graphics card releases from AMD, the firm is steadily ploughing a furrow in HPC and is showing off its latest wares for this market at the SC20 virtual tradeshow. Yesterday it took the wraps off the AMD Instinct MI100 Accelerator with the claim that it is "the world’s fastest HPC GPU accelerator for scientific workloads".

AMD says the Instinct MI100 is its first to use its new CDNA architecture which was built from the ground up to provide "a giant leap in compute and interconnect performance". Compared to previous AMD HPC accelerators the new MI100 offers a nearly 3.5x (FP32 Matrix) performance boost for HPC and a nearly 7x (FP16) performance boost for AI throughput, according to in-house testing.

Key technologies behind the MI100 GPU include; an all-new Matrix Core Technology with superior performance for Machine Learning, AMD Infinity Fabric Link Technology for 64GB/s CPU to GPU bandwidth and up to 276GB/s of peer-to-peer (P2P) bandwidth performance, PCIe Gen 4.0, and Ultra-Fast HBM2 memory - to deliver up to 11.5TFLOPS peak FP64 performance (or 23.1 TFLOPS peak FP32 performance).

AMD highlights the virtues of combining its award winning 2nd Gen Epyc processors with MI100 accelerators to deliver true heterogeneous compute capabilities for HPC and AI. It says that if you combine these two key innovative solutions and the open and portable AMD ROCm programming ecosystem the biggest challenges in HPC and AI can be tackled.

With regard to the usage of AMD's Instinct MI100 accelerator cards, the famous Oak Ridge Leadership Computing Facility has already been testing these GPUs as it was provided early access. Facility director Bronson Messer says that the MI100 has proved to deliver "significant performance boosts, up to 2-3x compared to other GPUs," in test platforms. In addition one also must factor in the energy efficiency savings for MI100.

 


HEXUS Forums :: 12 Comments

Login with Forum Account

Don't have an account? Register today!
But can it run Crysis?!


…(getting my coat)
WHAT?? A100 specs: FP32 Training 312 Tflops, INT8 Inference 1248 Tops & FP64 19.5 Tflops…..who is fooling who?. These data centre AI specs are becoming fuzzy.
lumireleon
WHAT?? A100 specs: FP32 Training 312 Tflops, INT8 Inference 1248 Tops & FP64 19.5 Tflops…..who is fooling who?. These data centre AI specs are becoming fuzzy.

Tflops doesn't necessarily translate between platforms/hardware, we only need to look at the ps5 and xbox series x for proof in that. It's entirely possible that in this exact field of usage that the current AMD card is better than the nvidia one.

In real terms, unless one us wins the lottery and knows how to code (or has access to them), we'll never really know or have any need to in all honesty.

edit: and which a100 are you referencing, the one with the article today with 4 cards inside or the card on it's own… they both have a100 in the name lol.
lumireleon
WHAT?? A100 specs: FP32 Training 312 Tflops, INT8 Inference 1248 Tops & FP64 19.5 Tflops…..who is fooling who?. These data centre AI specs are becoming fuzzy.

Those are the figures for the tensor cores which only have a specific compute focus whereas AMDs figures are for the general purpose compute which smokes the A100.
tharfcake
But can it run Crysis?!

Given the MI50 ended up as the Radeon VII after some tweaking, I would have to say “eventually” ;)

I am slightly surprised they don't slap a couple of DP ports on them and at least sell them as workstation video cards tbh. They would be awful value, but that doesn't seem to stop some people buying stuff.