diff --git a/projects/rccl-tests/doc/PERFORMANCE.md b/projects/rccl-tests/doc/PERFORMANCE.md index 21fef609af..942f054968 100644 --- a/projects/rccl-tests/doc/PERFORMANCE.md +++ b/projects/rccl-tests/doc/PERFORMANCE.md @@ -140,5 +140,6 @@ To obtain a bus bandwidth which should be independent of the number of ranks _n_ * AllGather : (_n_-1)/_n_ * Broadcast : 1 * Reduce : 1 +* AlltoAll: (_n_-1)/_n_ The bus bandwidth should reflect the speed of the hardware bottleneck : NVLink, PCI, QPI, or network.