NCCL issue 418: Multiple MPI ranks in the same GPU using Nvidia Multiprocess Service
repo
Fields
id
sources/mgpu-nccl-418-duplicate-device-mpsPublisher
NVIDIA/nccl
Kind
repo
Published
2020-11-11
Accessed
2026-10-03
notes
Captures the exact runtime signatures: 'NCCL WARN Duplicate GPU detected : rank 0 and rank 1 both on CUDA device ...' and "NCCL error a.cpp:106 'invalid usage'" from ncclCommInitRank.
Notes
Captures the exact runtime signatures: 'NCCL WARN Duplicate GPU detected : rank 0 and rank 1 both on CUDA device ...' and "NCCL error a.cpp:106 'invalid usage'" from ncclCommInitRank.
Cross-references
This record references no other records.