Failed run multiple nodes scripts on NVIDIA A100

[zhoujingyu13687306871]

Hello! I found that when I use openmpi, ucx, cuda, and gdrcopy to compile OpenCFD-SCU-v3.0, I can only run it on a single node. Running on multiple nodes will result in a segmentation fault. Has this program been tested on NVIDIA GPUs before?

[FR13ndSDP]

More investigation on Nvidia GPUs is needed.

[Changye112a]

你好，这套代码可以在Nvidia V100配置下多卡运行吗？