I currently need the ability the disable a certain number of compute units inside a GPU, is there any way of doing this with openCL 2.0? I was going to do it via clCreateSubDevices and essentially feed my program a sub device certain number of CU's via that call, and then the "other" device would just sit idle. Doing some research it seems that this API is not supported by AMD GPUs...is this still the case for Polaris GPUs as well? If so, any other way I could force certain number of CUs to not work?
Device fission is not currently supported on AMD GPUs and also, current implementation does not provide any such control to disable certain CUs in GPU.