This is different from CPU where L2 cache is shared by all GPU cores instead of L3 cache
There are 20 SMs for GTX 1080
Each SM contains 128 CUDA cores, 256 KB of register file capacity, a 96 KB shared memory unit, 48 KB of total L1 cache storage
32 threads/warp, 1 cuda thread = 1 cuda core
This is different from CPU where L2 cache is shared by all GPU cores instead of L3 cache
There are 20 SMs for GTX 1080
Each SM contains 128 CUDA cores, 256 KB of register file capacity, a 96 KB shared memory unit, 48 KB of total L1 cache storage
32 threads/warp, 1 cuda thread = 1 cuda core