NVIDIA/MatX

Health v2 · maintenance only

Adoption not part of the score

1444 stars · 126 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

MatX is a C++20 header-only numerical computing library providing NumPy-like tensor expressions for NVIDIA GPUs and multithreaded CPUs. It offers lazy expression evaluation, runtime kernel fusion, and easy frontends to CUDA and CPU libraries like cuBLAS, FFTW, and OpenBLAS.

Use cases

• write NumPy-style array code in C++ on GPUs
• fuse numerical pipelines to avoid kernel launches and memory traffic
• perform linear algebra, FFTs, convolutions, and reductions on tensors
• run the same tensor code on GPU or multithreaded CPU
• prototype GPU kernels without writing raw CUDA
• accelerate signal processing and scientific computing workloads

When to choose

• you need NumPy-like productivity with native C++ performance on NVIDIA GPUs
• you want a header-only library with no compilation step
• you want to switch between GPU and CPU executors without changing your algorithm
• you need high-performance tensor math for HPC or signal processing

When to avoid

• you need multi-vendor GPU support beyond NVIDIA
• you primarily work in Python and don't need C++ integration
• you need a full deep-learning framework with autodiff and training loops
• your project cannot use C++20

Facets

library · maturity active

gpu-computing math simulation machine-learning gpu-computing performance data-science developer-tools cpp windows cross-platform numerical-computing tensors numpy-like header-only c++20 hpc gpgpu lazy-evaluation linear-algebra fft algorithms cuda gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="NVIDIA/MatX")