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

NCCL Release 2.22.3

This is the NCCL 2.22.3 release notes. For previous NCCL release notes, refer to the NCCL Archives.

Compatibility

NCCL 2.22.3 has been tested with the following:

Key Features and Enhancements

This NCCL release includes the following key features and enhancements.

  • Lazy connection establishment for collective operations to speed up init and lower memory usage when not using all algorithms.

  • Accelerated Intra-node NVLink detection for faster init.

  • Init profiling, reporting time spent in different init phases.

  • Support for PCI p2p on split PCI switches.

  • Cost estimation API.

  • Net/IB: separate traffic class for fifo messages.

  • Net/IB: support for IB router.

  • Optimizations and fixes for device network offload (unpack).

  • Support ncclGroupStart/End for ncclCommAbort/Destroy.

  • Improved Tuner API.

  • Allow net and tuner plugins to be statically linked.

Fixed Issues

The following issues have been resolved in NCCL 2.22.3:

  • Disabled port fusion in heterogeneous systems.

  • Fixed NVLS graphs search for dual NICs.

  • Fixed crash with collnetDirect.

  • Fixed hang when nodes have different CPU types.

  • Fixed performance of registered send/recv operations.

  • Ensured ncclCommDestroy is non-blocking if ncclCommFinalize has been called.

  • Improved error reporting during SHM segment creation.

  • Improved support of various compilers.

Updating the GPG Repository Key

To best ensure the security and reliability of our RPM and Debian package repositories, NVIDIA is updating and rotating the signing keys used by apt, dnf/yum, and zypper package managers beginning on April 27, 2022. Failure to update your repository signing keys will result in package management errors when attempting to access or install NCCL packages. To ensure continued access to the latest NCCL release, please follow the updated NCCL installation guide.