General purpose GPU compute framework built on Vulkan to support 1000s of cross vendor graphics cards (AMD, Qualcomm, NVIDIA & friends). Blazing fast, mobile-enabled, asynchronous and optimized for advanced GPU data processing usecases. Backed by the Linux Foundation.

Vulkan compute for people

Gemma4.pas is a high-performance AI inference engine developed entirely in Delphi to run Google’s Gemma 4 model natively on Windows. Unlike most platforms, this project features zero external dependencies and utilizes Vulkan compute shaders to accelerate complex mathematical operations directly on the user's GPU.

A Highlevel Python Wrapper for Vulkan's Compute API

Cybervision can generate a 3D model from two photos of an object

Universal hardware acceleration bridge for Android. Proxies native GPU, DSP, and NPU capabilities to Linux Chroot environments via Unix Domain Sockets. Enables GPGPU, AI, and Compute offloading without driver limitation.

GTX-class post stack that approximates RTX-style lighting with Vulkan compute, screen-space techniques, and temporal accumulation. There is no hardware ray tracing, no DXR/VK ray tracing pipelines, and no DLSS dependency.

Vulkan Compute Library

Efficiently calculate the longest total stopping times of Collatz sequences.

Proof-of-concept prototype for Heterogeneous Virtual Threads (HVTs) — a proposed extension of Project Loom's Virtual Thread model to hardware accelerators. Companion to the paper "Heterogeneous Virtual Threads: Extending Project Loom for First-Class Accelerator Support in the JVM"

A 3D FDTD interactive simulator made with C++ and Vulkan

Unified memory ML inference engine for AMD APU (RDNA 3.5 / gfx1150) — Custom HIP kernels with 2.2x speedup over standard allocation

Vulkan Example 1, Vulkan Compute Matrix Transposition

Fractal generator/visualiser written as an experiment with multithreading and GPGPU