Skip to main content

A lone developer just open sourced a tool that could bring an end to Nvidia's AI hegemony — AMD financed it for months but abruptly ended its support. Nobody knows why

Web Hosting & Remote IT Support

Andrzej Janik, a developer working on a tool that allowed Nvidia's CUDA code to run on AMD and Intel GPUs without any modifications, has open sourced his creation after support for the project was dropped by both AMD and Intel.

Although ZLUDA received a major update last week, bringing it to version 3, its future going forward is now in doubt.

ZLUDA originally emerged in 2020 as a promising tool to enable Intel GPUs to run CUDA, the proprietary hardware-software ecosystem that underpins Nvidia's dominance. Janik lobbied for Intel to adopt it, but the company, who he worked for at the time, decided there was no business case for running CUDA applications on its GPUs and requested the project be taken down.

AMD picks up the baton

Janik subsequently left Intel and was contracted by AMD to continue ZLUDA development. However, AMD also concluded there was no business case for running CUDA applications on its GPUs and ended its support for the project. Janik was released from the contract and was able to bring ZLUDA back to the public domain. 

The current version of ZLUDA is significantly different from its 2020 iteration. Instead of being built on Intel's oneAPI and supporting Intel's GPUs, it is now based on AMD's ROCm solution and only supports Radeon GPUs. Janik stated that the project is more or less complete and will only receive updates for workloads he is personally interested in. 

The fact that neither Intel nor AMD are interested in making their GPUs compatible with the existing CUDA ecosystem is significant. Both companies appear to prefer competing directly with CUDA using their own open-source solutions, oneAPI and ROCm, despite CUDA's continued popularity in professional and datacenter graphics software. 

While AMD had been quietly funding ZLUDA for the past two years, the company decided to discontinue its support this year for unknown reasons. It’s possible that AMD wanted to avoid any possible lawsuits and so pulled out once the contract ended, meaning it couldn’t be directly tied to the project. 

Despite the lack of corporate backing, ZLUDA has shown promise in testing, with many CUDA software able to run on HIP/ROCm without any modifications. However, as Phoronix points out, it's not a fail-safe solution, with some features such as NVIDIA OptiX not being fully supported.

More from TechRadar Pro



via Hosting & Support

Comments

Popular posts from this blog

Microsoft, Google, and Meta have borrowed EV tech for the next big thing in data centers: 1MW watercooled racks

Web Hosting & Remote IT Support Liquid cooling isn't optional anymore, it's the only way to survive AI's thermal onslaught The jump to 400VDC borrows heavily from electric vehicle supply chains and design logic Google’s TPU supercomputers now run at gigawatt scale with 99.999% uptime As demand for artificial intelligence workloads intensifies, the physical infrastructure of data centers is undergoing rapid and radical transformation. The likes of Google, Microsoft, and Meta are now drawing on technologies initially developed for electric vehicles (EVs), particularly 400VDC systems, to address the dual challenges of high-density power delivery and thermal management. The emerging vision is of data center racks capable of delivering up to 1 megawatt of power, paired with liquid cooling systems engineered to manage the resulting heat. Borrowing EV technology for data center evolution The shift to 400VDC power distribution marks a decisive break from legacy sy...

The Apple Watch ban is lifted, on appeal – but the reprieve might only be temporary

Web Hosting & Remote IT Support The Apple Watch ban story has developed quickly over the last week and a bit, and there's now a new twist: the US Court of Appeals is putting a pause on the US sales and import ban while it reviews the case, which means the Apple Watch 9 and Apple Watch Ultra 2 can go back on sale for the time being. "We are thrilled to return the full Apple Watch lineup to customers in time for the new year," an Apple spokesperson told TechRadar. "We are pleased the US Court of Appeals for the Federal Circuit has stayed the exclusion order while it considers our request to stay the order pending our full appeal." The watches in question are now once again available from "select" Apple Stores, and will also be going on sale from the Apple website from 12pm PT / 3pm ET on Thursday, December 28 (that's 8pm in the UK, and early on December 29 in Australia). All Apple Stores should have stock by the weekend. As for how long t...

The Samsung Galaxy Ring could go into production as soon as next month

Web Hosting & Remote IT Support With the dust beginning to settle from the huge Samsung Unpacked 2023 event, we can turn our attention towards what Samsung might have planned next: and a smart ring seems to be in the company's near future. As per a report from South Korean outlet The Elec (via SamMobile ), mass production on a Samsung Galaxy Ring could begin as early as August, with a decision imminent on the schedule for getting the wearable manufactured and out to consumers. A full launch is slated for some point during 2024 though, rather than 2023. The nature of the device means that it'll need to clear several regulatory hurdles before it can go on sale and start tracking various vital statistics. An early 2024 launch would put the Galaxy Ring on a similar schedule to the Samsung Galaxy S24 – and it would therefore make sense to launch both gadgets at the same time, perhaps in January or February if Samsung follows its 2023 routine. The story so far Rumors ar...