TechPrufer
Get started
NVIDIA
NVIDIA logoNVIDIA · Feature trail

High-Performance Core Math Libraries in Python

2 posts · Jul 2026 – Aug 2026

What this trail covers

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, and cuTENSOR, with support for Universal Sparse Tensors (UST) and flexible installation options. The library allows for choice of memory and execution spaces, and provides both generic and specialized APIs for various use cases. Logging capabilities are integrated for tracing computational details and data flow. Composite operations are highlighted for their importance in performance optimization.

All NVIDIA trails
  1. CUDA Python 1.0: Stable APIs, One Foundation, Full Platform Access | NVIDIA Technical BlogCUDA Python 1.0 establishes a stable, unified foundation (`cuda.core`) for accessing the CUDA platform from Python. It introduces semantic versioning for key components (`cuda.core`, `cuda.compute`, `nvmath-python`), ensuring predictable API stability. `cuda.compute` exposes CCCL parallel algorithms as host-callable functions, and `cuda.bindings` provides 1:1 coverage of CUDA C APIs. This unified foundation improves interoperation between libraries like CuPy, PyTorch, and NVIDIA's own math/communication libraries, and provides easier access to advanced CUDA features like green contexts.
  2. Run High-Performance Core Math at Scale with NVIDIA nvmath-python | NVIDIA Technical BlogThis post announces the general availability of nvmath-python v1.0, detailing its design and capabilities for accelerating math operations across CPU, GPU, and distributed multi-node systems. It highlights the library's Pythonic abstraction over CUDA-X and NVPL math libraries, its fast and flexible installation, its role as a complement to existing array libraries like NumPy, CuPy, and PyTorch, and its support for various memory and execution spaces. The post also distinguishes between generic and specialized APIs, demonstrates logging capabilities for tracing computational details and data flow, and explains the importance of composite operations for performance.