Stacking Up AMD MI200 versus Nvidia A100 Compute Engines
nextplatform.com
Stacking Up AMD MI200 versus Nvidia A100 Compute Engines
1–10 of 39 posts
Re: Stacking Up AMD MI200 versus Nvidia A100 Compute Engines
#24.9x faster FP64, but only 1.2x faster FP16. FP64 is totally irrelevant for machine learning. A100 has been available on AWS for over a year already, while MI200 won't be available at all until next year. Nvidia's A100 successor should be out next year too. And Nvidia's software stack puts AMD's to shame. AMD needs to do a lot better if they want to start being competitive in ML.
Re: Stacking Up AMD MI200 versus Nvidia A100 Compute Engines
#34.9x faster FP64, but only 1.2x faster FP16. FP64 is totally irrelevant for machine learning. A100 has been available on AWS for over a year already, while MI200 won't be available at all until next year. Nvidia's A100 successor should be out next year too. And Nvidia's software stack puts AMD's to shame. AMD needs to do a lot better if they want to start being competitive in ML.
FP64 is very relevant for the kind of physics simulations that supercomputers are often built for. However the software stack is as well, and there I agree - it’s AMD’s main weak point.
Re: Stacking Up AMD MI200 versus Nvidia A100 Compute Engines
#4I find it interesting they’ve focused so much silicon on improving FP64 - in my mind that means they’re targeting physics simulations and other more traditional HPC workloads more than deep learning. I think that’s a smart thing on their part, because Nvidia really has a chokehold on the deep learning field right now with the A100 and CUDA/CUDNN software stack. I also find it interesting (and a good sign) that their recent supercomputer deals include 100 million for software development for ROCm/HIP.
However, considering how robust CUDA is compared to ROCm, I feel like all Nvidia would need to do to take the HPC market back completely would be to get close to AMD’s current FP64 performance. I don’t think anyone would buy AMD if prices were comparable and FP64 performance was anywhere in the ballpark. It’ll be very interesting to see Nvidia’s new cards, hopefully next year.
Re: Stacking Up AMD MI200 versus Nvidia A100 Compute Engines
#54.9x faster FP64, but only 1.2x faster FP16. FP64 is totally irrelevant for machine learning. A100 has been available on AWS for over a year already, while MI200 won't be available at all until next year. Nvidia's A100 successor should be out next year too. And Nvidia's software stack puts AMD's to shame. AMD needs to do a lot better if they want to start being competitive in ML.
Isn't the training done at FP32 and quantized aftweards?
Not that this would make this GPU more appealing. The FP32 performance isn't great either.
The memory bandwidth seems pretty good though.
Re: Stacking Up AMD MI200 versus Nvidia A100 Compute Engines
#64.9x faster FP64, but only 1.2x faster FP16. FP64 is totally irrelevant for machine learning. A100 has been available on AWS for over a year already, while MI200 won't be available at all until next year. Nvidia's A100 successor should be out next year too. And Nvidia's software stack puts AMD's to shame. AMD needs to do a lot better if they want to start being competitive in ML.
If you mean integrating the hardware stack to existing open-source software stacks, the yes, Nvidia has the clear advantage. What Nvidia has accomplished is not trivial but it’s not an impossible feat.
Re: Stacking Up AMD MI200 versus Nvidia A100 Compute Engines
#74.9x faster FP64, but only 1.2x faster FP16. FP64 is totally irrelevant for machine learning. A100 has been available on AWS for over a year already, while MI200 won't be available at all until next year. Nvidia's A100 successor should be out next year too. And Nvidia's software stack puts AMD's to shame. AMD needs to do a lot better if they want to start being competitive in ML.
Machine learning isn't the only useful thing to run on "GPU"s (these can't do graphics workloads anymore; they're basically just vector processors).
AMD isn't attempting to complete on f16 perf. They're completing where Nvidia's perf is abysmal: f64.
Having programmed AMDGPUs at a lower level than HIP/Rocm, they are actually much better than Nvidia (and in fact I'm able to do cool things like pcie large bar/p2p even on RDNA1 GPUs) in terms of flexibility. The HIP/CUDA API (nevermind OpenCL) doesn't do them justice, CUDA's enormous ecosystem and vender lock-in notwithstanding.
Re: Stacking Up AMD MI200 versus Nvidia A100 Compute Engines
#8I find it interesting they’ve focused so much silicon on improving FP64 - in my mind that means they’re targeting physics simulations and other more traditional HPC workloads more than deep learning. I think that’s a smart thing on their part, because Nvidia really has a chokehold on the deep learning field right now with the A100 and CUDA/CUDNN software stack. I also find it interesting (and a good sign) that their…
Supercomputer market is still there. Intel Sapphire rapids owes its existence to the next US nuke design (to be made on Aurora supercomputer.)
Re: Stacking Up AMD MI200 versus Nvidia A100 Compute Engines
#94.9x faster FP64, but only 1.2x faster FP16. FP64 is totally irrelevant for machine learning. A100 has been available on AWS for over a year already, while MI200 won't be available at all until next year. Nvidia's A100 successor should be out next year too. And Nvidia's software stack puts AMD's to shame. AMD needs to do a lot better if they want to start being competitive in ML.
Isn't the training done at FP32 and quantized aftweards? Not that this would make this GPU more appealing. The FP32 performance isn't great either. The memory bandwidth seems pretty good though.
There is a way to train with mixed fp16/fp32 precision.
Re: Stacking Up AMD MI200 versus Nvidia A100 Compute Engines
#10I find it interesting they’ve focused so much silicon on improving FP64 - in my mind that means they’re targeting physics simulations and other more traditional HPC workloads more than deep learning. I think that’s a smart thing on their part, because Nvidia really has a chokehold on the deep learning field right now with the A100 and CUDA/CUDNN software stack. I also find it interesting (and a good sign) that their…
Supercomputer market is still there. Intel Sapphire rapids owes its existence to the next US nuke design (to be made on Aurora supercomputer.)
I don't think so. The new ISA that is in the SPR is mostly about deep learning: it supports int8 and bfloat16 (https://fuse.wikichip.org/news/3600/the-x86-advanced-matrix-...). You can emulate higher precision using bfloat16 (https://arxiv.org/abs/1904.06376), but I have not seen this used in the wild.