GPU Acceleration

Questions and remarks about code_saturne usage
Forum rules
Please read the forum usage recommendations before posting.
Post Reply
StandardRANSUser001
Posts: 45
Joined: Tue Apr 09, 2024 3:26 am

GPU Acceleration

Post by StandardRANSUser001 »

Hi Everybody,

There seems to be a lot of development effort going into the GPU capabilities of CS. Is CS8.3 capable of GPU acceleration now?

If these capabilities are ready, could somebody please provide instructions on how to set up a case to benefit from an Nvidia HPC GPU?

Best regards,
Sean Hanrahan
Yvan Fournier
Posts: 4206
Joined: Mon Feb 20, 2012 3:25 pm

Re: GPU Acceleration

Post by Yvan Fournier »

Hello,

This is still work in progress, and usable on some code paths, not on others. But is it working on quite a few basic cases.
There is still room for many optimizations, and in any case you need a large enough mesh for it to be worthwhile, but this will be further improved by v9.0 release in June. Feedback is always welcome, though it is a tad early for high expectations.
As of the currrent state of things, for the BUNDLE test case (from saturne-open-cases), we get comparable performance on a single H100 as on a single node with 48 Xeon cores of ou main cluster, and should be able to improve upon this. Some operators are faster on CPu, others are faster on GPU.

I updated the INSTALL.md (install guide) file on the master branch on GitHub today, with a few examples, and the same is valid for v8.3.

Best regards,

Yvan
Post Reply