NVIDIA Deep Learning NCCL Documentation
Release Notes (PDF) - v2.20.5 - Last updated March 6, 2024

NCCL Release 2.1.2

Key Features and Enhancements

This NCCL release includes the following key features and enhancements.
  • New algorithms for improved latency communication
  • RoCE support

Using NCCL 2.1.2

Ensure you are familiar with the following notes when using this release.
  • The NCCL 2.x API is different from NCCL 1.x. Some porting may be needed for NCCL 1.x applications to work correctly. Refer to the migration documentation in the NCCL Developer Guide.

Known Issues

  • If NCCL returns an error code, set the environment variable NCCL_DEBUG to WARN to receive an explicit error message.
  • Using multiple processes in conjunction with multiple threads to manage the different GPUs may in some cases cause ncclCommInitRank to fail while establishing IPCs (cudaIpcOpenMemHandle). This problem does not appear when using only processes or only threads.
  • NCCL uses CUDA 9 cooperative group launch by default, which may induce increased latencies in multi-threaded programs. See the NCCL_LAUNCH_MODE knob in the NCCL Developer Guide to restore the original behavior.

Fixed Issues

  • NCCL now uses CUDA 9 cooperative groups to launch the CUDA kernels, fixing a long-standing issue with CUDA and NCCL operations being interleaved and potentially causing hangs.