How do I write a shader for Kompute?

Kompute uses GLSL shaders compiled to SPIR-V. You can write raw GLSL code as strings in your code or use provided tools to compile shaders, with examples in the README showing shader integration for tensor operations.

Is Kompute better than CUDA for machine learning?

Kompute offers cross-vendor support and mobile compatibility via Vulkan, while CUDA is NVIDIA-specific with a more mature ecosystem. Choose Kompute for hardware diversity or mobile targets, and CUDA for deep NVIDIA integration and libraries.

Can Kompute run on integrated graphics?

Yes, if the integrated GPU supports Vulkan. Kompute is designed to work with any Vulkan-compatible device, including integrated graphics from Intel or AMD, as mentioned in its cross-vendor features.

How to set up Kompute for Android development?

Use the Android NDK wrapper included in the repository, and follow the end-to-end examples in the documentation. It involves integrating Kompute into Android Studio projects with CMake, as detailed in the mobile-enabled section.

What's the performance difference between Kompute and OpenCL?

Kompute leverages Vulkan, which can offer lower overhead and better parallelism on modern GPUs, but performance varies by hardware. The README cites examples like parallel queue operations, but benchmarks depend on specific use cases.

How does memory management work in Kompute?

Kompute uses explicit memory management with tensors that handle GPU and host memory. You need to sync data using operations like OpTensorSyncDevice and OpTensorSyncLocal, as shown in the asynchronous processing examples.

Open-Awesome

Vulkan Kompute

Apache-2.0C++v0.9.0

A general-purpose GPU compute framework built on Vulkan for cross-vendor graphics cards, enabling high-performance data processing and machine learning.

Visit Website GitHub

2.5k stars194 forks0 contributors

What is Vulkan Kompute?

Kompute is a general-purpose GPU compute framework built on the Vulkan API, designed to enable high-performance data processing and machine learning across a wide range of graphics cards from vendors like AMD, Qualcomm, and NVIDIA. It provides abstractions for GPU memory management, asynchronous operations, and shader execution, simplifying the development of GPU-accelerated applications.

Target Audience

Developers and researchers working on GPU-accelerated machine learning, data processing, or game development who need cross-vendor compatibility and fine-grained control over GPU resources.

Value Proposition

Kompute offers a balance between low-level Vulkan control and high-level usability, with support for both C++ and Python, mobile deployment, and asynchronous processing, making it a versatile choice for advanced GPU computing without vendor lock-in.

Overview

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.

Use Cases

Best For

Implementing machine learning models with GPU acceleration on cross-vendor hardware
Building mobile applications with on-device GPU-accelerated data processing
Developing game engines that require custom GPU compute shaders
Parallelizing heavy GPU workloads across multiple queue families
Prototyping GPU-accelerated algorithms in Python with seamless C++ integration
Running GPU compute tasks on Android devices via the NDK

Not Ideal For

Projects requiring out-of-the-box machine learning models without custom shader development
Teams exclusively targeting NVIDIA GPUs where CUDA's mature ecosystem and libraries are preferred
Applications needing minimal setup and quick deployment without Vulkan SDK dependencies
Developers seeking a purely high-level API without any GPU programming knowledge

Pros & Cons

Pros

Cross-Vendor GPU Support

Works with AMD, Qualcomm, NVIDIA, and other Vulkan-compatible cards, enabling deployment across diverse hardware without vendor lock-in, as highlighted in the README's key features.

Asynchronous & Parallel Processing

Supports GPU family queues for parallel execution and asynchronous operations, optimizing performance for heavy workloads through sequences and queue allocation, as detailed in the architectural overview.

Mobile and Desktop Versatility

Optimized for Android NDK with examples for mobile GPU acceleration, and works on desktop platforms, making it suitable for edge computing and cross-device applications.

Flexible Language Bindings

Provides both a high-level Python module for rapid prototyping and a low-level C++ SDK for performance tuning, supported by interactive notebooks and comprehensive documentation.

Cons

Shader Development Required

Users must write and compile custom GLSL shaders for compute tasks, adding complexity compared to frameworks with pre-built kernels, as seen in the examples where raw shader code is necessary.

Vulkan Dependency and Setup

Relies on the Vulkan SDK and compatible drivers, which can be a barrier in environments without Vulkan support or where installation is cumbersome, as noted in the build system overview.

Limited High-Level Abstractions

While it abstracts Vulkan boilerplate, it still requires low-level GPU programming knowledge and lacks ready-to-use algorithms for common tasks, making it less accessible for quick prototyping.

Frequently Asked Questions

Related Projects

ncnn

ncnn is a high-performance neural network inference framework optimized for the mobile platform

Cross-platform, graphics API agnostic, "Bring Your Own Engine/Framework" style rendering library.

A multi-platform library for OpenGL, OpenGL ES, Vulkan, window and input

Stars15,079

Forks5,889

Last commit2 months ago

MoltenVK

MoltenVK is a Vulkan Portability implementation. It layers a subset of the high-performance, industry-standard Vulkan graphics and compute API over Apple's Metal graphics framework, enabling Vulkan applications to run on macOS, iOS and tvOS.

Stars5,666

Forks515

Last commit4 days ago

Community-curated · Updated weekly · 100% open source

Found a gem we're missing?

Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.

Submit a project Star on GitHub