We’re excited to announce the nvmath-python 1.0 General Availability release.
This milestone brings:
- Stable APIs for building on NVIDIA core math libraries from Python
- Multi-GPU, multi-node support across the stack: cuFFTMp, cuBLASMp, and cuDSS with continued performance improvements
- New cuSOLVERMp-backed LU solver support
Since the early beta releases, nvmath-python has grown from single-GPU FFT, matmul, and device APIs into a broader Python interface for dense, sparse, tensor, distributed, and device-side math workflows. Thank you to everyone who tested, filed issues, and shared feedback along the way. We’re excited to see what you build with v1.0!
More about this release: