ROCm Communication Collectives Library (RCCL)

Stories

AMD adds GPU-initiated SDMA copy engines to RCCL and rocSHMEM

Added a GPU-initiated SDMA backend, enabled with NCCL_GIN_TYPE=7

AMD now lets a running GPU kernel hand large data transfers to dedicated System Direct Memory Access (SDMA) copy engines, instead of having compute units (CUs) execute every load and store. The feature is in RCCL, AMD's ROCm Communication Collectives Library, and rocSHMEM, its GPU networking library. Both are part of the ROCm software stack, and the feature is available in version 10.1.

Updated 1 hour ago