OneCCL Versions Save

What's New

• Performance improvements for scaleup for all message sizes for AllReduce, Allgather, and Reduce_Scatter.
• Optimizations also include small message sizes that appear in inference apps.
• Performance improvements for scaleout for Allreduce, Reduce, Allgather, and Reduce_Scatter.
• Optimized memory usage of oneCCL.
• Support for PMIx 4.2.6.
• Bug fixes.

Removals

• oneCCL 2021.12 removes support for PMIx 4.2.2

This update provides bug fixes to maintain driver compatibility for Intel® Data Center GPU Max Series.

This update addresses stability issues with distributed Training and Inference workloads on Intel® Data Center GPU Max Series.

• Added point to point blocking communication operations for send and receive.

• Performance optimizations for Reduce-Scatter.

• Improved profiling with Intel® Instrumentation and Tracing Technology (ITT) profiling level.

• Improved scaling efficiency of the Scaleup algorithms for ReduceScatter
• Optimized performance of oneCCL scaleup collectives by utilizing the new embedded Data Streaming Accelerator in Intel® 4th Generation Xeon Scalable Processors (formerly code-named Sapphire Rapids)

• Optimizations across the board including improved scaling efficiency of the Scaleup algorithms for Alltoall and Allgather • Add collective selection for scaleout algorithm for device (GPU) buffers

• Provides optimized performance for Intel® Data Center GPU Max Series utilizing oneCCL. • Enables support for Allreduce, Allgather, Reduce, and Alltoall connectivity for GPUs on the same node.

Intel® oneAPI Collective Communications Library 2021.7.1 has been updated to include functional and security updates. Users should update to the latest version as it becomes available.

• Bug fixes

• Added support for Intel® ITT (Intel® Instrumentation and Tracing Technology) profiling
• Added WSL2 (Windows Subsystem for Linux 2) support
• Moved dependency check for Level Zero from compile-time to runtime
• Bug fixes