Intel nixes its next-gen AI GPU but still has plans to take on Nvidia

Fair to say Intel’s GPU plans don’t always go to, er, plan. During Intel’s latest earnings call for highly-remunerated bean counters, the company’s new interim co-CEO let slip that its upcoming next-gen AI GPU has effectively been cancelled.

Known as Falcon Shores, it was supposed to replace Intel’s Gaudi 3 chip. But no more. “Based on industry feedback, we plan to leverage Falcon Shores as an internal test chip only without bringing it to market,” Michelle Holthaus said.

Instead, Intel has switched its focus to the generation of AI GPU that comes next, Jaguar Shores. “This will support our efforts to develop a system-level solution at rack scale with Jaguar Shores to address the AI data center,” Holthaus said.

Our understanding is that the problem with Intel’s AI GPUs is that they’re all a bit bare bones as an offering. Where Nvidia will sell you a fully built up rack of GPUs with fancy interconnects, Intel doesn’t have the same mature and complete product. Oh, and there’s the minor matter of CUDA, Nvidia’s software interface that allows developers to code in existing computer languages and port right over to Nvidia’s GPUs.

“One of the things that we’ve learned from Gaudi is it’s not enough to just deliver the silicon. We need to be able to deliver a complete rack scale solution, and that’s what we’re going to be able to do with Jaguar Shores,” Holthaus explained.

Without necessarily wanting to dance on Intel’s grave with this one, let’s just say the company doesn’t exactly have the best record when it comes to GPUs. Who can forget Larrabee, the chip with zillions of tiny x86 cores that was cancelled back in 2010 but might have put Intel and indeed its recently departed CEO Pat Gelsinger, who championed the Larrabee project, on a completely different path.

Intel’s Arc GPUs haven’t exactly been a roaring success, either, even if its latest Intel Arc B580 board is a clear step forward. So, it would be a brave call to predict that Jaguar Shores will soon be chasing Nvidia’s H100 and B200 GPUs out of AI data centers any time soon.

In the meantime, ye olde Linux patch text strings have outed what appears to be references to Intel’s next-gen Celestial gaming GPU in high-performance discrete format, which tallies with recent comments by Intel’s main graphics rep Tom Petersen that the hardware for Celestial is basically done and dusted.

So, Intel’s GPU efforts rumble on. Eventually, maybe one of them will really land. We genuinely hope so, it would be great for competition.