Repolode
Submit
KomputeProject

KomputeProject/kompute

A C++ and Python GPU compute framework built on Vulkan that runs general-purpose workloads on AMD, Qualcomm, NVIDIA and other vendors' GPUs.

Open-source alternative to:CUDA

View on GitHubWebsite
Stars
2,581
Forks
202
Open issues
79
Last commit
3 months ago

What is it?

Kompute is a general-purpose GPU compute framework that uses Vulkan as its backend, so the same code can run on graphics cards from different vendors, including AMD, Qualcomm and NVIDIA. It wraps the boilerplate of the Vulkan compute API in a manager, tensors, sequences and algorithms, letting developers move data to the GPU, dispatch shaders and read results back with far less setup code.

The framework offers a C++ SDK for low-level control and a Python package for interactive experimentation. Operations can be recorded into sequences and submitted asynchronously, and work can be spread across multiple GPU queue families. It is designed to coexist with existing Vulkan applications rather than hide the Vulkan API.

Kompute also targets mobile use, with dynamic loading of the Vulkan library on Android and an NDK wrapper for its C++ headers. The project includes examples for logistic regression, Android apps and Godot game integration.

Key features

  • 01Cross-vendor GPU support through Vulkan, including AMD, Qualcomm and NVIDIA hardware.
  • 02C++ SDK for low-level optimization and a Python package for interactive use.
  • 03Asynchronous execution and parallel operations across GPU queue families.
  • 04Explicit management of GPU and host memory via tensors.
  • 05Android NDK wrapper and dynamic Vulkan loading for mobile builds.
  • 06Bring-your-own-Vulkan design that works alongside existing Vulkan code.

Use cases

  • Running machine learning or numerical workloads on GPUs without CUDA, such as on AMD or Qualcomm hardware.
  • Adding GPU compute to an existing Vulkan renderer or game engine.
  • Bringing on-device GPU inference or data processing to Android apps.
  • Prototyping GPU algorithms in Python before moving them to C++.

Things to know

  • Requires a Vulkan-capable GPU and driver, since it is built on Vulkan.
  • The README badge lists Python support only for versions 3.7 to 3.9.

In collections

Android apps & phone tools

Topics

#vulkan#gpu-computing#gpgpu#machine-learning#cross-platform#c-plus-plus#python#mobile#cpp#deep-learning

FAQ

What is kompute?
A C++ and Python GPU compute framework built on Vulkan that runs general-purpose workloads on AMD, Qualcomm, NVIDIA and other vendors' GPUs. Kompute is a general-purpose GPU compute framework that uses Vulkan as its backend, so the same code can run on graphics cards from different vendors, including AMD, Qualcomm and NVIDIA. It wraps the boilerplate of the Vulkan compute API in a manager, tensors, sequences and algorithms, letting developers move data to the GPU, dispatch shaders and read results back with far less setup code.
Is kompute free to use? What license does it have?
kompute is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is kompute still maintained?
Updated within 6 months. The last commit to kompute was on Aug 15, 2026.
What can I use kompute for?
Running machine learning or numerical workloads on GPUs without CUDA, such as on AMD or Qualcomm hardware. Adding GPU compute to an existing Vulkan renderer or game engine. Bringing on-device GPU inference or data processing to Android apps. Prototyping GPU algorithms in Python before moving them to C++.
Where can I get kompute?
The source code and installation instructions are on GitHub: https://github.com/KomputeProject/kompute

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects