ZLUDA Open-Source Library Revived: Aims to Enable CUDA-Based AI Workloads on Multiple GPU Architectures

ZLUDA's Revival and New Direction

ZLUDA, the open-source "code porting" library, has been resurrected with a new focus on AI and machine learning workloads. Originally designed to support Intel GPUs on NVIDIA's software stack, ZLUDA has undergone significant changes in its development journey 1

1

The project's original developer, Andrzej Janik, has announced that ZLUDA is back in development under the sponsorship of an anonymous backer. This new iteration aims to break down exclusivity barriers in AI software stacks by enabling CUDA-based AI workloads to run on GPUs from multiple vendors 2

2

Multi-GPU Compatibility and AI Focus

The revived ZLUDA is being developed with a focus on multi-GPU compatibility, particularly for AI workloads. This means the library will be designed to work with various GPU architectures, including those from AMD and NVIDIA. The project is shifting its focus from professional workloads to AI and machine learning applications 1

1

Key developments include:

1. Support for AI/ML libraries such as Llama.cpp, PyTorch, and TensorFlow
2. Reworking NVIDIA code paths for compatibility with other GPU vendors
3. Initial testing with AMD's RDNA GPUs
4. Planned support for RDNA1+ architectures and ROCm 6.1+ compute stack 1

   1

Development Timeline and Open-Source Collaboration

Janik estimates that it will take approximately a year for the new ZLUDA code to reach a stable state capable of effectively handling AI/ML workloads across multiple GPUs. The project will remain open-source, welcoming contributions from the community as it evolves 2

2

Related Stories

ZLUDA Project Advances: Bringing NVIDIA's CUDA to Non-NVIDIA GPUs

Solo Developer Successfully Runs CUDA Applications on AMD GPUs Using ZLUDA and HIP Translation Layer

NVIDIA CUDA Applications Now Compatible with AMD GPUs Thanks to SCALE Toolkit

French startup ZML launches free software to speed AI inference across Nvidia, AMD, and rival chips

Potential Impact on AI Hardware Landscape

If successful, ZLUDA could significantly impact the AI hardware landscape by allowing different GPU architectures to leverage each other's capabilities. This could potentially lead to more optimal end results in AI and machine learning applications 1

1

Anonymous Sponsorship and Future Revelations

While ZLUDA now has financial backing, the sponsor has chosen to remain anonymous for the time being. Speculation suggests that the sponsor may be a large entity interested in running AI workloads at scale with multi-GPU vendor support. The developer has indicated that the identity of this 'stealth' sponsor is expected to be revealed later, which could provide more insight into ZLUDA's future direction and support 2

2

As ZLUDA enters this new phase of development, it has the potential to become a game-changer in the AI compute landscape, particularly for AMD's compute portfolio. The project's success could lead to a more open and flexible environment for AI and machine learning workloads across different GPU architectures.