6 ms·

Boosting Computational Fluid Dynamics Performance with AMD MI300X

47 points by latchkey 2y ago

thomasfedb 2y ago
Our team has access to multiple systems that either have MI250Xs or H100s. Getting stuff to work with AMD/ROCm is substantially more effort than the NVIDIA/CUDA experience. Some of this is lack of groundwork/engineering by packages or system administrators, but it seems a decent amount is the relative lack of effort by AMD to make things work well OOTB.
- BoingBoomTschak 2y ago
  The real question is: is this from lack of effort or simply from NVidia's headstart? Will it get better?
  - noch 2y ago
    > Will it get better? It won't, not in any way that will make AMD approximately competitive with Nvidia. AMD, unlike Nvidia, seems unable to prioritize developers. Here's a summary of last week's charlie-fox when the TinyGrad team attempted to get 2 MI300s for on-premises testing and was rebuffed by an AMD representative. https://x.com/dehypokriet/status/1879974587082912235 https://x.com/dehypokriet/status/1879974587082912235
  - musicale 2y ago
    Both, really. It will get better, but Nvidia seems to be creating CUDA libraries for all kinds of applications, so the moat is constantly widening/deepening.
- latchkey 2y ago
  Installing ROCm is easy and well documented [0]. Anush (AMD VP of AI software) has had a fire lit under his butt after the recent SemiAnalysis article [1] and is actively taking feedback on improving the experience. If you have specific things you'd like to see, I'm more than happy to forward them onto him (contact in my profile). [0] https://rocm.docs.amd.com/en/latest/ https://rocm.docs.amd.com/en/latest/ [1] https://semianalysis.com/2024/12/22/mi300x-vs-h100-vs-h200-benchmark-part-1-training/ https://semianalysis.com/2024/12/22/mi300x-vs-h100-vs-h200-b...
lukasb 2y ago
warning - no pretty videos in this post
gymbeaux 2y ago
Now if only I could purchase one* *I realize enterprise “GPUs” are difficult to purchase as an individual whether they’re AMD or Nvidia, but AMD might be well-served to mimic their RX 480 strategy (“make a competitive mid-range GPU, distribute it through several board partners, and sell it at rock-bottom prices to get it to as many customers as possible”). If there’s a compelling reason to use AMD hardware over Nvidia, the software support will come. As an extreme example, if one could purchase an MI300X for $50 a pop, surely FAANG and others would invest time and effort into the software side to leverage the cost savings vs Nvidia, who is unquestionably price-gouging due to their monopolistic hold on the enterprise GPU market.
- mpreda 2y ago
  It seems AMD has to choose between monetizing the GPUs right now by selling them at the maximum effective price, or forgoing that gain and selling the GPUs at cost and increasing GPU market share in the future. The choice between short-term gain and long-term gain.
  - mpreda 2y ago
    And the splitting into CDNA and RDNA comes from the same direction: market segmentation, to allow much higher prices for the CDNA data-center GPUs, while keeping the gamer-focused RDNA GPUs affordable for mere mortals. Of coures this backfires by making the powerful GPUs not available for mostly anybody anymore to experiment on. For example this blog post, about how great MI300X is. Really, what do I care -- I'm not a billionaire.
    - tormeh 2y ago
      They’re unifying the architectures. AMD will move to UDNA for both gaming and data center. The next graphics cards after RDNA4 will be UDNA. Makes sense given how ML-heavy graphics has become.
      - kouteiheika 2y ago
        The point is they shouldn't have done it in the first place. It was obvious right from the start it's a bad idea, except maybe for temporarily boosting short term profits. The whole AMD AI/ML strategy feels like this - prioritize short term profits and completely shoot themselves in the foot in the long term.
JorgeGT 2y ago
It's tangential, but this is the first time I've seen Fluent installed by simply decompressing a tar, instead of executing their big installer.
semi-extrinsic 2y ago
This looks like a really stupid benchmark, no? 4 million cells in a standard aerodynamic simulation isn't enough work to even saturate the host CPUs on these systems, never mind the GPUs which have far "worse" roofline models from a CFD algorithm point of view. I would not be surprised at all if these benchmarks ran faster if you removed the GPUs completely.
- shihab 2y ago
  That's the smallest of 4 experiments. It goes upto 140 million cells, where MI300X retains similar performance advantage of around 10% over Nvidia's H100.
  - semi-extrinsic 2y ago
    Yeah, but the lack of comparison to a CPU-only system is suspicious. (As is the exaggerated plot without ticks and labels on the y-axis.) I've never seen a non-reactive incompressible flow simulation get substantial speedup on GPUs. There are well understood fundamental reasons why this is the case.
    - shihab 2y ago
      I helped develop a hydro solver coupled with radiation at Los Alamos this summer. We observed from 7x upto 15x performance improvement on a single GPU compared to 64-core intel CPU. [1] Now granted, the flops to byte ratio for this program might be better than an avg fluid simulator. Also, our performance tanked when we moved to multi-node system. But I am aware of underlying reasons behind the scalibility issues and they don't feel like problems that can't be overcome. [1] https://github.com/lanl/HARD/tree/main https://github.com/lanl/HARD/tree/main
      - semi-extrinsic 2y ago
        Nice, interesting solver! But yes, compressible/density-based solvers work a lot better on GPUs than incompressible/pressure-based solvers. When 80% of the runtime is spent solving a Poisson equation, GPU is not worth it. Especially if you do the comparison on equivalent cost basis, i.e. "what is the walltime difference if I run on a $60k all-CPU cluster versus a $60k GPU cluster". Or in terms of cloud compute cost / HPC allocation spend.
KeplerBoy 2y ago
Those are some sketchy bar charts. If the bar doesn't start at y=0 and you don't even have axis ticks, it's meaningless and just a colorful blob on your slide.

47 points by latchkey 2y ago

Our team has access to multiple systems that either have MI250Xs or H100s. Getting stuff to work with AMD/ROCm is substantially more effort than the NVIDIA/CUDA experience. Some of this is lack of groundwork/engineering by packages or system administrators, but it seems a decent amount is the relative lack of effort by AMD to make things work well OOTB.

BoingBoomTschak 2y ago
The real question is: is this from lack of effort or simply from NVidia's headstart? Will it get better?
- noch 2y ago
  > Will it get better? It won't, not in any way that will make AMD approximately competitive with Nvidia. AMD, unlike Nvidia, seems unable to prioritize developers. Here's a summary of last week's charlie-fox when the TinyGrad team attempted to get 2 MI300s for on-premises testing and was rebuffed by an AMD representative. https://x.com/dehypokriet/status/1879974587082912235 https://x.com/dehypokriet/status/1879974587082912235
- musicale 2y ago
  Both, really. It will get better, but Nvidia seems to be creating CUDA libraries for all kinds of applications, so the moat is constantly widening/deepening.
latchkey 2y ago
Installing ROCm is easy and well documented [0]. Anush (AMD VP of AI software) has had a fire lit under his butt after the recent SemiAnalysis article [1] and is actively taking feedback on improving the experience. If you have specific things you'd like to see, I'm more than happy to forward them onto him (contact in my profile). [0] https://rocm.docs.amd.com/en/latest/ https://rocm.docs.amd.com/en/latest/ [1] https://semianalysis.com/2024/12/22/mi300x-vs-h100-vs-h200-benchmark-part-1-training/ https://semianalysis.com/2024/12/22/mi300x-vs-h100-vs-h200-b...

lukasb 2y ago

warning - no pretty videos in this post

gymbeaux 2y ago

Now if only I could purchase one* *I realize enterprise “GPUs” are difficult to purchase as an individual whether they’re AMD or Nvidia, but AMD might be well-served to mimic their RX 480 strategy (“make a competitive mid-range GPU, distribute it through several board partners, and sell it at rock-bottom prices to get it to as many customers as possible”). If there’s a compelling reason to use AMD hardware over Nvidia, the software support will come. As an extreme example, if one could purchase an MI300X for $50 a pop, surely FAANG and others would invest time and effort into the software side to leverage the cost savings vs Nvidia, who is unquestionably price-gouging due to their monopolistic hold on the enterprise GPU market.

mpreda 2y ago
It seems AMD has to choose between monetizing the GPUs right now by selling them at the maximum effective price, or forgoing that gain and selling the GPUs at cost and increasing GPU market share in the future. The choice between short-term gain and long-term gain.
- mpreda 2y ago
  And the splitting into CDNA and RDNA comes from the same direction: market segmentation, to allow much higher prices for the CDNA data-center GPUs, while keeping the gamer-focused RDNA GPUs affordable for mere mortals. Of coures this backfires by making the powerful GPUs not available for mostly anybody anymore to experiment on. For example this blog post, about how great MI300X is. Really, what do I care -- I'm not a billionaire.
  - tormeh 2y ago
    They’re unifying the architectures. AMD will move to UDNA for both gaming and data center. The next graphics cards after RDNA4 will be UDNA. Makes sense given how ML-heavy graphics has become.
    - kouteiheika 2y ago
      The point is they shouldn't have done it in the first place. It was obvious right from the start it's a bad idea, except maybe for temporarily boosting short term profits. The whole AMD AI/ML strategy feels like this - prioritize short term profits and completely shoot themselves in the foot in the long term.

JorgeGT 2y ago

It's tangential, but this is the first time I've seen Fluent installed by simply decompressing a tar, instead of executing their big installer.

semi-extrinsic 2y ago

This looks like a really stupid benchmark, no? 4 million cells in a standard aerodynamic simulation isn't enough work to even saturate the host CPUs on these systems, never mind the GPUs which have far "worse" roofline models from a CFD algorithm point of view. I would not be surprised at all if these benchmarks ran faster if you removed the GPUs completely.

shihab 2y ago
That's the smallest of 4 experiments. It goes upto 140 million cells, where MI300X retains similar performance advantage of around 10% over Nvidia's H100.
- semi-extrinsic 2y ago
  Yeah, but the lack of comparison to a CPU-only system is suspicious. (As is the exaggerated plot without ticks and labels on the y-axis.) I've never seen a non-reactive incompressible flow simulation get substantial speedup on GPUs. There are well understood fundamental reasons why this is the case.
  - shihab 2y ago
    I helped develop a hydro solver coupled with radiation at Los Alamos this summer. We observed from 7x upto 15x performance improvement on a single GPU compared to 64-core intel CPU. [1] Now granted, the flops to byte ratio for this program might be better than an avg fluid simulator. Also, our performance tanked when we moved to multi-node system. But I am aware of underlying reasons behind the scalibility issues and they don't feel like problems that can't be overcome. [1] https://github.com/lanl/HARD/tree/main https://github.com/lanl/HARD/tree/main
    - semi-extrinsic 2y ago
      Nice, interesting solver! But yes, compressible/density-based solvers work a lot better on GPUs than incompressible/pressure-based solvers. When 80% of the runtime is spent solving a Poisson equation, GPU is not worth it. Especially if you do the comparison on equivalent cost basis, i.e. "what is the walltime difference if I run on a $60k all-CPU cluster versus a $60k GPU cluster". Or in terms of cloud compute cost / HPC allocation spend.

KeplerBoy 2y ago

Those are some sketchy bar charts. If the bar doesn't start at y=0 and you don't even have axis ticks, it's meaningless and just a colorful blob on your slide.