# CuPy

CuPy is an open-source array library for GPU-accelerated computing with Python. It is open source and self-hosted, installable on Docker Hub.

Canonical page: https://ourlocalhost.com/app/6450/cupy

## Description

CuPy is an open-source array library for GPU-accelerated computing with Python. CuPy utilizes CUDA Toolkit libraries including cuBLAS, cuRAND, cuSOLVER, cuSPARSE, cuFFT, cuDNN and NCCL to make full use of the GPU architecture.

## Platform availability

Available on 0 of 10 platforms.

## Health score

65/100 (Good), based on 4 of 5 factors.
- Maintained: 100 (last commit 18 days ago)
- Popular: 82 (12,390 GitHub stars)
- Easy to install: 25 (0 install platforms, install notes, Docker image)
- Security check: 15 (2 serious findings, 6 minor, CI, tests, OpenSSF Scorecard 3.8/10)

## Security check

2 serious findings: A GitHub workflow can be triggered by outsiders to run their code with the project's secrets; Uses numpy 2.3., which has a known critical vulnerability.
Automated checks on commit 65cfd1a, 15 Oct 2026. Not an audit.

## Links

- Docker Hub: https://hub.docker.com/r/cupy/cupy
- Documentation: https://docs.cupy.dev/en/stable/
- GitHub: https://github.com/cupy/cupy
- Official website: https://cupy.dev

## Common questions

### How can I verify the CUDA runtime actually used by an installed CuPy build?

Import CuPy and print cupy.cuda.get_local_runtime_version() to see the CUDA runtime shared library actually present. The default version call may report the version CuPy was built with, not the version present on your machine.

### Do I need to install cuTENSOR or NCCL for basic CuPy to work?

No, cuTENSOR and NCCL are optional GPU libraries for extra CuPy features. Install them with matching CUDA extras only if a recipe needs them; without them, basic CuPy still works but those specific functions are missing.

### Where should I report a CuPy problem that comes from conda-forge?

Report it to cupy-feedstock. The maintainers can investigate whether it is a packaging issue in the conda-forge recipe or a real issue in CuPy.

### Why might the first CuPy run be slower?

The first run can be slowed by CUDA kernel caching. The referenced thread asks for ways to reduce that initial slowdown while staying in high-level Python.

## Alternatives

- Calcpad: Calcpad is a lightweight mathematical platform for cloud and desktop.
- Julia: Julia is a powerful tool used to write computer programmes.
- MATLAB: MATLAB is a numerical computing platform for data analysis, algorithm development, mathematical modelling, and visualisation, available as…
- R (programming language): R is a free software environment for statistical computing and graphics.
- SageMath: SageMath is open-source mathematical software that combines many specialised tools into a single, easy-to-use system.
- fxSolver: Online math solver, equation editor and database, graphing calculator and science/engineering problem helper.
- wxMaxima: wxMaxima is a document based interface for the computer algebra system Maxima.

---
Source: Our Local Host (https://ourlocalhost.com/app/6450/cupy), the self-hosted and open-source app directory.
