EC/CUDA One-shot Kernel with Cooperative Launch

This feature improves GPU collective performance by utilizing the CUDA cooperative launch feature. It enables the use of a single CUDA kernel for CUDA operations in UCC GPU collectives.

This feature can activated by enabling the UCC environment variable UCC_EC_CUDA_USE_COOPERATIVE_LAUNCH as follows:

Copy
Copied!
            

UCC_EC_CUDA_USE_COOPERATIVE_LAUNCH=1

© Copyright 2023, NVIDIA. Last updated on Oct 5, 2023.