加载中
正在获取最新内容，请稍候...
ZLUDA: Run CUDA Applications on AMD and Intel GPUs
ZLUDA is a compatibility layer that allows applications written for NVIDIA's CUDA™ platform to run on different GPUs, specifically targeting AMD and Intel graphics hardware. It aims to bridge the gap between CUDA software and non-NVIDIA ecosystems.
Rust
Added on 2025年5月13日11,315
Stars721
ForksRust
LanguageProject Introduction
Summary
ZLUDA is an open-source project acting as a CUDA compatibility layer, allowing software designed for NVIDIA GPUs to execute on other vendors' hardware like AMD and Intel, thereby expanding the accessibility of CUDA-dependent applications.
Problem Solved
CUDA applications are typically locked into NVIDIA's hardware ecosystem, preventing users with AMD or Intel GPUs from utilizing CUDA-dependent software. ZLUDA provides an alternative by enabling these applications to run on non-NVIDIA platforms.
Core Features
CUDA API Translation
Provides a runtime translation layer for CUDA API calls to equivalent non-NVIDIA GPU APIs like HIP (for AMD) or potentially SYCL/Level Zero (for Intel).
Performance Focus
Designed with performance in mind, aiming to achieve execution speeds comparable to native implementations on the target hardware.
Targeted Compatibility
Currently supports specific versions of CUDA Toolkit and focuses on core functionalities required by various applications.
Tech Stack
C++
HIP (for AMD)
SYCL / Level Zero (for Intel)
Use Cases
ZLUDA can be used in various scenarios where CUDA-dependent software needs to be executed on non-NVIDIA hardware:
Running Scientific Simulations
Details
Executing scientific simulation software (e.g., GROMACS, LAMMPS if ZLUDA supports their CUDA backend) on clusters equipped with AMD Instinct accelerators or Intel Max Series GPUs.
User Value
Access to powerful simulation capabilities on diverse hardware, potentially leveraging existing non-NVIDIA infrastructure.
AI/ML Model Inference
Details
Testing or running certain AI/ML inference workloads that are distributed as CUDA binaries, on consumer or professional AMD/Intel graphics cards.
User Value
Enabling access to AI/ML applications for users without NVIDIA hardware, broadening the reach of such software.
Developer Tooling and Libraries
Details
Using developer tools or libraries that have a CUDA dependency but are desired for use on a machine with an AMD or Intel GPU for development or testing purposes.
User Value
Increased flexibility in development environments, allowing developers to work with CUDA-dependent tools regardless of their primary GPU vendor.
Recommended Projects
You might be interested in these projects
immortalwrtimmortalwrt
ImmortalWrt is an open-source embedded operating system based on OpenWrt, specifically tailored and optimized for users in mainland China, offering enhanced features, stability, and compatibility.
C
79832304
overleafoverleaf
A web-based, collaborative LaTeX editor designed to simplify document creation and teamwork for academic writing, reports, presentations, and more.
JavaScript
152801575
coturncoturn
coturn is a free open source implementation of TURN and STUN servers. It is used to facilitate NAT traversal for real-time communications applications like WebRTC, VoIP, and online gaming.
C
123012107