Allow GPU kernel to use only specific SM

I am an introductory person who only has a basic knowledge of CUDA and GPU.

Now I want to do “assigning a kernel to a specific SM (one or multiple) so that multiple kernels can be performed within GPUs at the same time”.

But it’s very hard to find any related data.

Is it possible to do this? If it can, how can I do it?

CUDA doesn’t provide you any explicit way to do this.

1 Like