Process per node vs process per GPU #2
Description
HiCCL (and CommBench) works only for single-process-per-GPU mapping. It would be great to support single-processes-per-multiple-GPUs so that it can be used in more number of applications.
{{ message }}