NVIDIA CUDA 13.3 has been released with several notable additions. CUDA Python 1.0 reaches stable status, providing a supported way to leverage CUDA in Python applications for AI, data science, and scientific computing. For C++ developers, CUDA Tile programming model is now available. The release also introduces CompileIQ, a compiler auto-tuning framework delivering up to 15% speed-ups on kernels like GEMM and attention. Additional updates include a Numba CUDA MLIR back-end, math library updates, C++23 support in NVCC and NVRTC, and mmap() support.
2 Impressions