I’m new to CUDA and confusing with the relationship between the number of CUDA cores and the size of warps.
It’s known that a GT200 GPU has 8 CUDA cores in a SM, and there are 4 threads run concurrently in a CUDA core, so there are 32 threads in a warp.
Now, i have a GTX460, which has 48 CUDA cores in a SM. Why the warp size is also 32?
Are there any idle CUDA cores in a SM?