kompute

Why we included this project

Kompute gives you a way to write and run compute kernels on Vulkan without locking yourself into one GPU vendor. If your workload involves data processing or machine learning and needs GPU speed, but you don't want to adopt a heavyweight deep learning framework, this is a vendor-neutral option with both a C++ SDK and a Python module. Teams already using Vulkan can drop kernels into an existing application without swapping out their rendering setup. The framework handles the tedious parts of GPU and host memory ownership and async queue dispatch, so you focus on the kernel logic instead of driver plumbing. That makes it a good fit for custom kernels doing data transforms, simulation, or ML pre and post processing across desktop and mobile.