I am running the nvbandwidth test on a 8*A100 nvswitch server. My question is, when GP0 reads from GPU1 using cudaMemcpy() and P2P is enabled, can cudaMemcpy() use both the link of pci and nvlink? that is split data into two parts and transfer data via pcie and nvlink separately?
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| How does “cudaMemcpyPeer” implement? | 3 | 1299 | February 6, 2024 | |
| How to close nvlink connections | 1 | 532 | December 16, 2019 | |
| about the nvlink between two gpus | 4 | 1071 | April 3, 2019 | |
| Can Unified Memory Migration use NVLink? | 2 | 714 | October 12, 2021 | |
| NVLINK | 3 | 1988 | May 5, 2018 | |
| How to use p2p over PCIe or NVLink between different containers on the same node? | 0 | 37 | August 16, 2024 | |
| NVLink and Cache Levels | 0 | 544 | July 19, 2023 | |
| Passing messages from one GPU to another | 2 | 483 | October 16, 2020 | |
| HPL test using NVLINK | 1 | 1292 | March 11, 2023 | |
| The cudaMemcpy speed between two A100Gpus is slow | 1 | 654 | December 22, 2021 |