4 ms·
The GEMM instructions are sent to Tensor Cores by the scheduler. The cutlass library is the open source version of cublas: https://github.com/NVIDIA/cutlass ht
by rubatuga 5y ago
The GEMM instructions are sent to Tensor Cores by the scheduler. The cutlass library is the open source version of cublas:
https://github.com/NVIDIA/cutlass https://github.com/NVIDIA/cutlass
- floatboth 5y agoSeems like you can leverage those via VK_NV_cooperative_matrix…