Add support for CUDA graphs.
Fuse BCM Gen4 switches to avoid suboptimal performance on some platforms. Issue #439.
Fix bootstrap issue caused by connection reordering.
Fix CPU locking block.
Improve CollNet algorithm.
Improve performance on DGX A100 for communicators with only one GPU per node.
This commit is contained in:
Sylvain Jeaugey
2021-04-12 16:00:11 -07:00
والد 911d61f214
کامیت a46ea10583
43فایلهای تغییر یافته به همراه2687 افزوده شده و 1244 حذف شده
+2 -2
مشاهده پرونده
@@ -1,6 +1,6 @@
##### version
NCCL_MAJOR := 2
NCCL_MINOR := 8
NCCL_PATCH := 4
NCCL_MINOR := 9
NCCL_PATCH := 6
NCCL_SUFFIX :=
PKG_REVISION := 1