For my experiment, I want to know is there any way to disable part multiprocessors in CUDA ? For example, I want to use less than 8 multiprocessors instead of 16 multiprocessors
This would be a useful tool for scalability testing, but it’d have to be supported at pretty low driver level.
There is driver-level support for changing core and memory clocks of the GPU though.
Thank you for your reply.
Does it have Linux version of NVIDIA NVAPI?