2 ms·
Nice presentation. None of what was said is specific to CUDA. It’s the same thing in DirectCompute HLSL. Only the terminology is different: CUDA thread blocks
by Const-me 4y ago
Nice presentation.
None of what was said is specific to CUDA. It’s the same thing in DirectCompute HLSL. Only the terminology is different: CUDA thread blocks are called thread groups in DirectCompute, warps are called waves, shared memory is called “group shared”, etc.
Probably applies to others too (Vulkan, Metal, etc.) albeit I never used these for HPC-like stuff, not 100% sure.