NVIDIA Deep Learning NCCL Documentation
Release Notes (PDF) - v2.21.5 - Last updated June 18, 2024

NCCL Release 2.0.5

Key Features and Enhancements

This NCCL release includes the following key features and enhancements.
  • NCCL 2.0.5 provides support for intra-node and inter-node communication.
  • NCCL optimizes intra-node communication using NVLink, PCI express, and shared memory.
  • Between nodes, NCCL implements fast transfers over sockets or InfiniBand verbs.
  • GPU-to-GPU and GPU-to-Network direct transfers, using the GPU Direct technology, are extensively used when the hardware topology permits it.

Using NCCL 2.0.5

Ensure you are familiar with the following notes when using this release.
  • The NCCL 2.0 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.
  • NCCL 2.0.5 has the new configuration file support. The NCCL environment variables can now be set in ~/.nccl.conf and /etc/nccl.conf.
  • Values defined in ~/.nccl.conf take precedence over ones in /etc/nccl.conf.
  • The syntax for each line of the NCCL configuration file is <NCCL_VAR_NAME>=<VALUE>.

Known Issues

  • If NCCL returns any error code, set the environment variable NCCL_DEBUG to WARN to receive an explicit error message.
  • RoCE support is experimental.
  • 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.