CuPy: NumPy & SciPy for GPU


Keywords
cublas, cuda, cudnn, cupy, curand, cusolver, cusparse, cusparselt, cutensor, gpu, nccl, numpy, nvrtc, nvtx, python, rocm, scipy, tensor
License
MIT
Install
pip install cupy-cuda111==12.3.0

Documentation

CuPy : NumPy & SciPy for GPU

pypi Conda GitHub license Gitter Twitter

Website | Install | Tutorial | Examples | Documentation | API Reference | Forum

CuPy is a NumPy/SciPy-compatible array library for GPU-accelerated computing with Python. CuPy acts as a drop-in replacement to run existing NumPy/SciPy code on NVIDIA CUDA or AMD ROCm platforms.

>>> import cupy as cp
>>> x = cp.arange(6).reshape(2, 3).astype('f')
>>> x
array([[ 0.,  1.,  2.],
       [ 3.,  4.,  5.]], dtype=float32)
>>> x.sum(axis=1)
array([  3.,  12.], dtype=float32)

CuPy also provides access to low-level CUDA features. You can pass ndarray to existing CUDA C/C++ programs via RawKernels, use Streams for performance, or even call CUDA Runtime APIs directly.

Installation

Pip

Binary packages (wheels) are available for Linux and Windows on PyPI. Choose the right package for your platform.

Platform Architecture Command
CUDA 11.x (11.2+) x86_64 / aarch64 pip install cupy-cuda11x
CUDA 12.x x86_64 / aarch64 pip install cupy-cuda12x
ROCm 4.3 (experimental) x86_64 pip install cupy-rocm-4-3
ROCm 5.0 (experimental) x86_64 pip install cupy-rocm-5-0

Note

To install pre-releases, append --pre -U -f https://pip.cupy.dev/pre (e.g., pip install cupy-cuda11x --pre -U -f https://pip.cupy.dev/pre).

Conda

Binary packages are also available for Linux and Windows on Conda-Forge.

Platform Architecture Command
CUDA x86_64 / aarch64 / ppc64le conda install -c conda-forge cupy

If you need to use a particular CUDA version (say 11.8), you can do conda install -c conda-forge cupy cuda-version=11.8.

Note

If you encounter any problem with CuPy installed from conda-forge, please feel free to report to cupy-feedstock, and we will help investigate if it is just a packaging issue in conda-forge's recipe or a real issue in CuPy.

Docker

Use NVIDIA Container Toolkit to run CuPy container images.

$ docker run --gpus all -it cupy/cupy

Resources

License

MIT License (see LICENSE file).

CuPy is designed based on NumPy's API and SciPy's API (see docs/source/license.rst file).

CuPy is being developed and maintained by Preferred Networks and community contributors.

Reference

Ryosuke Okuta, Yuya Unno, Daisuke Nishino, Shohei Hido and Crissman Loomis. CuPy: A NumPy-Compatible Library for NVIDIA GPU Calculations. Proceedings of Workshop on Machine Learning Systems (LearningSys) in The Thirty-first Annual Conference on Neural Information Processing Systems (NIPS), (2017). [PDF]

@inproceedings{cupy_learningsys2017,
  author       = "Okuta, Ryosuke and Unno, Yuya and Nishino, Daisuke and Hido, Shohei and Loomis, Crissman",
  title        = "CuPy: A NumPy-Compatible Library for NVIDIA GPU Calculations",
  booktitle    = "Proceedings of Workshop on Machine Learning Systems (LearningSys) in The Thirty-first Annual Conference on Neural Information Processing Systems (NIPS)",
  year         = "2017",
  url          = "http://learningsys.org/nips17/assets/papers/paper_16.pdf"
}