We have also described the structure of an efficient GEMM in our talk at the GPU Technology Conference 2018.

• CUTLASS: Software Primitives for Dense Linear Algebra at All Levels and Scales within CUDA
• Developing CUDA Kernels to Push Tensor Cores to the Absolute Limit on NVIDIA A100
• Accelerating Convolution with Tensor Cores in CUTLASS
• Accelerating Backward Data Gradient by Increasing Tensor Core Utilization in CUTLASS
• CUTLASS: Python API, Enhancements, and NVIDIA Hopper