Achieve up to 75% Performance Improvement for Communication Intensive HPC Applications with NVTAGS – NVIDIA Technical Blog News and tutorials for developers, data scientists, and IT admins 2025-03-26T22:01:23Z http://www.open-lab.net/blog/feed/ Logan Herche <![CDATA[Achieve up to 75% Performance Improvement for Communication Intensive HPC Applications with NVTAGS]]> http://www.open-lab.net/blog/?p=33648 2022-08-21T23:52:02Z 2021-06-23T21:29:15Z Many GPU-accelerated HPC applications spend a substantial portion of their time in non-uniform, GPU-to-GPU communications. Additionally, in many HPC systems,...]]> Many GPU-accelerated HPC applications spend a substantial portion of their time in non-uniform, GPU-to-GPU communications. Additionally, in many HPC systems,...

Many GPU-accelerated HPC applications spend a substantial portion of their time in non-uniform, GPU-to-GPU communications. Additionally, in many HPC systems, different GPU pairs share communication links with varying bandwidth and latency. As a result, GPU assignment can substantially impact time to solution. Furthermore, on multi-node / multi-socket systems, communication performance can degrade��

Source

]]> 0 ���˳���97caoporen����