Closed dmargala closed 3 years ago
I also updated the "how to build gpu conda env" instructions (resolves #52)
For typical benchmarking, I am just using one node with 4 gpus. I will add another section with instructions that typical configuration on corigpu and dgx.
Multinode is usually just a matter of changing the number of requested nodes in the salloc/sbatch setting.
I added instructions for build a conda env on dgx (unfortunately openmpi is not built for the latest cuda there, it's on the dgx todo and should be done in the next few weeks.)
I also added another doc with configurations for running typical single frame and 30 frame benchmarks.
LGTM, you can merge when you're ready. Thanks for the additional details.
This PR updates the profiling documentation (resolves #53). I've added instructions for profiling spex using mpi + gpu. It looks like some of the other existing instructions are also out of date so I will update those as well.