Closed zhiyang-fu closed 1 month ago
Hi Zhiyang,
Did you figure out how to use the multiple GPUs at the end? The mpirun command you used works well for me when you have the multiple GPUs in the same computer. You need to input a host file with individual IPs if you want to access different computers. THe output line "on processor "**" " does not seem correct.
Best regards,
Andreu
Hi Andreu,
Thank you so much for the reply. I was able to use multiple GPUs in the end yet with some limitations. The mistake was that I was using anaconda's mpirun (which somehow is the default). After I switched to the system's mpirun, I can use multiple GPUs except that I could not use all the GPUs of the server. It seems that the program complaints that one of the GPUs is connected to a display:
==> CUDA: GPU #0 is connected to a display and the CUDA driver would limit the kernel run time. Skipping this GPU!!
Is there a way to bypass the warning and ask the program to not skip this GPU? In fact, I do not think the server is connected to any display.
Best regards, Zhiyang
Hi Andreu,
I would like to run MCGPU using multiple gpus on a single machine using the following command line. However, the program ended up running on 4 cpus yet the same gpu.
mpirun -n 4 ./MC-GPU_v1.3.x MC-GPU_v1.3.in
I checked the output file, it printed the following relevant lines. The number of processors is still 1. Do you have any thoughts on what caused the problem? Thanks.
` CUDA SIMULATION IN THE GPU