Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Support to NVIDIA Grace-Hopper Systems #405

Open
ccchen057 opened this issue Sep 7, 2024 · 3 comments
Open

Support to NVIDIA Grace-Hopper Systems #405

ccchen057 opened this issue Sep 7, 2024 · 3 comments

Comments

@ccchen057
Copy link

Is there any plan to support or test on NVIDIA Grace-Hopper systems (Arm Architecture and NVIDIA H100)?

@jcandy
Copy link
Member

jcandy commented Sep 9, 2024

@sfiligoi Any thoughts?

@sfiligoi
Copy link
Contributor

sfiligoi commented Sep 9, 2024

It should just work.
There is nothing special about Grace Hopper... just another "NVIDIA-GPU-based system"
All you need is access to the HW and a recent compiler (ideally, NVIDIA HPC SDK)

@sfiligoi
Copy link
Contributor

The only potential "problem" we may encounter is a performance one;
since the CPU and GPU can access each other memories, OpenACC/OpenMP may have a slightly different semantics.

e.g. We currently assume we can initialize a buffer on the CPU and then copy that buffer to the GPU, and get optimal performance.
If the buffer would instead reside wherever it was "first touched", that may result in expensive remote data accesses if we are not careful.

Again, not a showstopper, but something to consider if we get significant G-H compute resources in the future.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants