NVIDIA's high-performance CUDA C++ template library providing reusable GEMM and linear algebra components plus Python DSL for researchers and engineers writing optimized GPU kernels.