diff --git a/src/collectives/device/common.h b/src/collectives/device/common.h index a53ae58da5..6e563b811c 100644 --- a/src/collectives/device/common.h +++ b/src/collectives/device/common.h @@ -11,7 +11,12 @@ #include "collectives.h" #include "devcomm.h" +#if defined(__gfx940__) #define COLL_UNROLL 4 +#else +#define COLL_UNROLL 2 +#endif + #define NCCL_MAX_DEV_ARITY (NCCL_MAX_TREE_ARITY-1) // Using balanced tree instead of split tree #define __syncwarp()