Any setting(enivronment variable etc) to not all GPU's in Multi-GPU hardware?


I am using Nvidia Teslas2050 GPU(contains 4 M2050 GPU’s) connected to x86_64 machine.
When I try to access any specific GPU from inside the container, all the other GPU’s are polled.

Please let me know if there is any setting of environmental variable etc to avoid the Nvidia Driver
polling all the GPU’s.

Thanks in advance


CUDA_VISIBLE_DEVICES might do what you want. Set it to 0 or 0,1 or 1,0 or something like that–comma-separated list of CUDA device IDs.

Thanks for the reply.

But it didn’t help.