AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

2.10.3-1

Browse Source 
Add support for bfloat16.
Add ncclAvg reduction operation.
Improve performance for aggregated operations.
Improve performance for tree.
Improve network error reporting.
Add NCCL_NET parameter to force a specific network.
Add NCCL_IB_QPS_PER_CONNECTION parameter to split IB traffic onto multiple queue pairs.
Fix topology detection error in WSL2.
Fix proxy memory elements affinity (improve alltoall performance).
Fix graph search on cubemesh topologies.
Fix hang in cubemesh during NVB connections.
This commit is contained in:
Ke Wen
2021-07-08 14:12:04 -07:00
parent 3fec2fa5ee
commit 7e51592129
52 changed files with 3496 additions and 2469 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graph/topo.h
+2 -3
View File
@@ -9,12 +9,11 @@
#include "graph.h"
#include "core.h"
#include <sched.h>
#define LOC_WIDTH 5000.0
#define SM60_NVLINK_WIDTH 18.0
#define SM70_NVLINK_WIDTH 21.0
#define SM80_NVLINK_WIDTH 21.0
#define SM70_NVLINK_WIDTH 22.0
#define SM80_NVLINK_WIDTH 22.0
#define SM86_NVLINK_WIDTH 12.0
#define PCI_WIDTH 12.0 // PCI Gen3 x16
#define QPI_WIDTH 6.0