The CUDA warp matrix functions utilize NVIDIA GPU's "tensor cores" functional units to enable matrix multiply-add (MMA) operations in CUDA kernels.
Questions tagged [cuda-wmma]
1 questions
1
vote
1 answer
Warp Matrix-Multiply functions - are single-precision multiplicands supported?
In the CUDA Programming guide, v11.7, section B.24.6. Element Types & Matrix Sizes, there's a table of supported type combinations, in which the multiplications are either sub-single-precision floating point types, or double - never `float . But -…
einpoklum
- 118,144
- 57
- 340
- 684