When I run a CUDA or OPENCL program, can I know how many sm and sp the gpu used for this program?
any software can do that?
gpu kernels are usually called in many thousand threads which occupy 100% of gpu
you mean gpu will use its all the cores?
yes
so if I only use only 1 block in cuda program. The gpu will use the the cuda cores in one sm or use all the cuda cores in gpu?
each thread block work inside a single SM, so a kernel with one thread block will use only cores of a single SM
multiple thread blocks are divided evenly between all SMs
thank u very much.