nv-legate/cupynumeric

End of life; no longer maintained. Final public release: v26.06.01.

What it solves

It addresses the difficulty of scaling NumPy-based workflows to high-performance computing environments. It allows users to run existing NumPy code on GPUs and distributed systems without requiring significant code changes.

How it works

cuPyNumeric implements the NumPy API on top of the Legate framework, enabling array computing to scale from a single CPU to thousands of GPUs across multiple nodes.

Who it’s for

It is designed for researchers and developers working on large-scale data analysis, complex simulations, and machine learning who need to scale their computations across multiple GPUs and nodes.

Highlights

  • Implements the NumPy API for high-performance computing.
  • Scales from a single CPU to thousands of GPUs across multiple nodes.
  • Requires little to no code changes to existing NumPy workflows.
  • Built on top of the Legate framework.

Related

  • Project
    cupy/cupy

    CuPy is a NumPy and SciPy-compatible array library that enables GPU-accelerated numerical computing for Python on NVIDIA CUDA and AMD ROCm platforms.

  • Project
    NVIDIA/cuda-python

    A collection of Python packages providing both high-level idiomatic and low-level binding access to NVIDIA's CUDA platform for GPU-accelerated computing.

  • Project
    inducer/pycuda

    PyCUDA is a Python wrapper for Nvidia's CUDA API that enables high-performance parallel computation on GPUs with automatic memory management and C++ speed.

  • Project
    lupinemachines/lupine

    LUPINE is a Docker‑based GPU‑over‑IP bridge that forwards CUDA driver calls from a CPU‑only machine to a remote GPU server, letting you run CUDA programs (e.g., PyTorch, nvidia‑smi) as if the GPU were locally installed.

  • Project
    NVIDIA/cudnn-frontend

    NVIDIA’s cuDNN Frontend is a header‑only C++ library plus Python bindings that simplify use of the cuDNN Graph API. It offers a unified graph abstraction, auto‑tuning, and a suite of open‑source high‑performance kernels (Flash‑Attention, fused GEMM‑+‑activation, sparse attention, etc.) for Hopper and Blackwell GPUs, supporting FP16/BF16/FP8. Install via pip or by including the header, and integrate directly with PyTorch or CUDA Graphs.