The UXL Foundation, a group of major chip industry players, expects that its alternative to Nvidia Corp.’s CUDA technology will become “mature” by year’s end. 

Reuters reported the group’s development schedule today. The technology the UXL Foundation’s members are building could make it easier to move workloads from Nvidia chips to rival processors. In the long term, that may create more competition for the graphics card maker.

The software interface through which developers interact with Nvidia’s chips is known as CUDA. It provides the programming building blocks necessary to run a neural network on the company’s graphics cards. Additionally, CUDA includes tools that ease related tasks such as distributing an AI model across multiple GPUs.

CUDA-powered applications can only run on Nvidia silicon. Rewriting such software for AI accelerators from rival chipmakers is possible, but requires a significant amount of time and effort. The UXL Foundation launched last September to ease the task. 

The consortium is backed by Intel Corp., Samsung Electronics Co. Ltd., Qualcomm Inc. and several other major chipmakers. Reuters reported today that the UXL Foundation is working to expand its membership with additional chip industry players. It’s believed the group also hopes to draw major cloud providers.

So far, the UXL Group’s development efforts have focused on a software toolkit called OneAPI. Developed by Intel, the toolkit makes it possible to write AI applications that can be moved between different chips with relative ease. OneAPI is based on an earlier framework called SYCL that likewise focuses on easing application portability. 

Intel has extended the framework with several additional features, most notably a capability called SYCLomatic. It’s designed to convert software written for Nvidia’s CUDA into SYCL code capable of running on other companies’ AI chips. Vinesh Sukumar, Qualcomm’s head of AI and machine learning, told Reuters that “we’re actually showing developers how you migrate out from an Nvidia platform.”

Reducing the amount of effort, and thus the cost, involved in moving Nvidia-powered applications to rival chips could potentially make enterprises more likely to use those chips. In the long term, that may create more competition for Nvidia’s market-leading graphics cards.

Reuters reported that oneAPI is “already usable.” The UXL Foundation reportedly hopes to build on the technology to create a “standard programming model of computing designed for AI.” 

So far, the effort has drawn technical contributions from both the consortium’s member companies and third parties. The UXL Foundation’s technical steering committee is expected to prepare the specifications for its AI programming model during the first half of 2024. The technology is expected to become mature by year’s end. 

Photo: Unsplash