KomputeProject/kompute

Health v2 · maintenance only

Adoption not part of the score

2558 stars · 198 forks observed · 2026-08-28

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

Kompute is a general-purpose GPU compute framework built on Vulkan that works across vendor GPUs (AMD, NVIDIA, Qualcomm, etc.) with both C++ and Python interfaces. It supports asynchronous and parallel GPU operations, mobile deployment via Android NDK, and is a Linux Foundation (LF AI & Data) hosted project.

Use cases

• run gpu compute shaders across amd nvidia and qualcomm gpus
• accelerate machine learning inference on mobile devices with vulkan
• process tensor data on the gpu from python
• integrate gpu compute into an existing vulkan application
• run parallel asynchronous gpu workloads in c++
• gpu acceleration in game engines like godot
• compute on android devices without cuda

When to choose

• you need cross-vendor GPU compute without CUDA or vendor lock-in
• you target mobile GPUs via Vulkan and the Android NDK
• you want a lightweight C++/Python layer over Vulkan compute with explicit memory management
• you already have a Vulkan application and want bring-your-own-Vulkan integration

When to avoid

• you need mature deep-learning training ecosystems like PyTorch or TensorFlow
• you only target NVIDIA GPUs and can use CUDA directly
• you need high-level prebuilt neural network layers rather than raw compute kernels
• your team cannot write or manage SPIR-V compute shaders

Facets

framework · maturity active

gpu-computing machine-learning sdk gpu-computing machine-learning deep-learning cross-platform mobile-development cpp python windows cross-platform vulkan gpgpu compute-shaders spirv mobile-gpu tensor-compute android linux macos mobile

Member repositories

For agents

markdown · JSON · MCP: product_card(name="KomputeProject/kompute")