Welcome to Item Finder Ville!

Jim Keller on AI, RISC-V, Tenstorrent’s Move to Edge IP

//php echo do_shortcode(‘[responsivevoice_button voice=”US English Male” buttontext=”Listen to Post”]’) ?>

Videography: Arthur Alvarez

Legendary CPU designer Jim Keller took over as CEO of AI chip company Tenstorrent at the beginning of 2023, after serving as the company’s CTO for two years. His history includes stints at Apple, Tesla and AMD. In recent years, Keller has become an outspoken supporter of RISC-V, and the burgeoning open-source ISA was a key topic for discussion during EE Times’ exclusive video interview.

“My belief is in the next 5 to 10 years, RISC-V will take over all the data centers,” Keller told EE Times, adding that this is especially true for scientific computing and HPC. He said supercomputing could dominate even faster.

Keller is a big believer in open source for both hardware and software; Tenstorrent intends to open-source its own AI software stack imminently.

“We wanted to do this last year, but we weren’t ready—our software stack was too messy and it needed to be partitioned the right way,” he said. “We aspire to be a company that open-sources software successfully.”

In a surprise move for a data center chip company, Tenstorrent recently licensed both its Tensix AI accelerator core IP and its Ascalon CPU core IP to LG Electronics. The Korean consumer electronics giant plans to use the Tensix IP at the embedded edge, in smart TVs and automotive chips. The two companies also plan to collaborate on future generations of RISC-V CPU, AI accelerator and video codec IP and/or chiplets. (LG spun out its own AI IP division in 2020).

Describing customer interest in Tenstorrent’s IP cores as a “pleasant surprise,” Keller said the company is still in business exploration mode.

“To be honest, we’re early in our business… we’re still small and we’re working out a lot of technology issues,” he said in our exclusive video interview. “As a technologist, I’m really interested in engagement with smart people, so if somebody smart says: I want open-source access to your hardware so I can program it the way I want, why would I say no to that?”

Keller’s view of the edge AI market is that alternative IP offerings so far have been too focused, and too hard to program.

“In the last two years, there’s been five pivots,” he said. “Is it image or language, is it inference or training? Is it a big model or a small model? Is it generative or not? Everybody who targeted an IP, the next model didn’t work. One of our premises from the founding was AI is going to evolve a lot, the differentiation between inference and training, language and image are going to be blurry, they’re going to move back and forth.”

Keller’s example, StableDiffusion, is part image model, part language model with the backwards pass of a training pipeline—in his words, “it looks like the kitchen sink—which IP does that run on?”

Watch the EE Times exclusive interview with Keller, above, for more on:

  • His strategy for Tenstorrent,
  • RISC-V in the data center and HPC,
  • His view of the competition, including hyperscalers’ own chips,
  • The role of chaos in innovation, and whether AI can innovate,
  • Neuromorphic computing, and
  • AI regulation.

We will be happy to hear your thoughts

Leave a reply

Compare items
  • Total (0)
Shopping cart