According to Reuters the UXL Foundation—a consortium of tech companies—aims to forge a suite of software and tools capable of powering various AI accelerator chips. The open-source initiative strives to ensure computer code operates on any device, irrespective of the underlying chip and hardware.
Qualcomm's head of AI and machine learning, Vinesh Sukumar, said: "We're showing developers how to transition away from an Nvidia platform.
Google's director and chief technologist of high-performance computing, Bill Magro, said the idea was to create an open ecosystem within machine learning frameworks, fostering productivity and hardware choice.
UXL's technical steering committee is on track to finalise technical specifications in the first half of this year. By year's end, engineers anticipate honing these details to a "mature" state.
The executives underscored the importance of laying a robust foundation that welcomes contributions from diverse companies and is adaptable to any chip or hardware.
UXL is set to engage cloud-computing behemoths like Amazon.com and Microsoft's Azure, alongside additional chip manufacturers.
Since its inception in September, UXL has received technical input from both members and external parties eager to leverage open-source technology. Intel's OneAPI is already operational, and the next phase involves creating a standard AI computing programming model.
UXL is dedicating its efforts to tackling pressing computing challenges monopolised by a handful of chipmakers, focusing on cutting-edge AI applications and high-performance computing. These initial steps align with the organisation's long-term ambition to attract a significant developer following to its platform.
In the future, UXL plans to extend support to Nvidia hardware and software.
Nvidia does not seem too concerned. Executive Ian Buck told Reuters: "The world is getting accelerated. New ideas in accelerated computing are emerging from the entire ecosystem, which will propel AI and the reach of accelerated computing."
The UXL Foundation's strategy is among numerous attempts to erode Nvidia's grip on AI-powering software. Venture capitalists and corporate investments have funnelled over €3.67 billion into 93 distinct initiatives, as per data curated by PitchBook for Reuters.
The drive to dethrone Nvidia via a perceived software vulnerability has intensified over the past year, with startups keen to puncture the firm's leadership securing just above €1.83 billion in 2023, a stark increase from €531.48 million the previous year, based on PitchBook figures.
Achieving success in Nvidia's shadow, particularly in AI data processing, is a feat few startups can claim. Nvidia's CUDA stands out as a robust software solution, continually expanding through contributions from both Nvidia and its developer community.
Jay Goldberg, CEO of D2D Advisory said a key point is the longstanding use of CUDA for 15 years, around which developers have constructed their code."