High-Performance Core Math Libraries in Python

posts

2026

This feature thread tracks the evolution of providing high-performance core math operations to Python developers, enabling them to leverage NVIDIA's accelerated math libraries (CUDA-X, NVPL) for CPU, GPU, and distributed systems. Initial efforts focused on bridging the gap between Python's scientific ecosystem and low-level GPU acceleration. Subsequent developments have introduced libraries like nvmath-python, offering a Pythonic abstraction layer over libraries such as cuFFT, cuBLASLt, cuDSS, cuSPARSE, cuTENSOR, and cuBLASMp. This includes advancements in universal sparse tensor (UST) support, generic and specialized APIs for various use cases, flexible installation options, and seamless integration with popular array libraries like NumPy, CuPy, and PyTorch. The focus is on enabling efficient execution across diverse hardware and distributed environments, with features like logging for computational tracing and amortization of autotuning costs.

2026

Run High-Performance Core Math at Scale with NVIDIA nvmath-python | NVIDIA Technical Blog

7/31/2026

This post introduces nvmath-python v1.0, a Pythonic abstraction layer over CUDA-X and NVPL math libraries. It enables Python users to access accelerated math operations on CPU, GPU, and distributed multi-node systems, integrating with NumPy, CuPy, and PyTorch. Key contributions include universal sparse tensor (UST) support via a domain-specific language, generic and specialized APIs for performance optimization, fast and flexible installation, and integration with Python's logging module for computational tracing. It also highlights the ability to amortize planning and autotuning costs and allows for custom kernel fusion.