Toronto startup Tenstorrent Inc. scored a major coup today with the news that legendary computer chip designer Jim Keller has signed on as its new president and chief technology officer.

Keller (pictured) is widely regarded as a “rock star” chip designer who created breakthrough processors at a number of leading tech firms, including Advanced Micro Devices Inc., Apple Inc. and Tesla Inc. Most recently he served as a senior vice president at Intel Corp., leaving that company last summer for “personal reasons.”

Keller started out at Digital Equipment Corp. in the 1990s, but it was at AMD that he really made a name for himself by helping to create that company’s Athlon K7 and K8 processors which enabled it to rival Intel in the computer server chip market. He joined Apple Inc. in 2008, working on the A series processors used in early iPhone models, before rejoining AMD in 2012 to help design its Zen architecture.

He also had a short stint at Tesla before landing in 2018 at Intel, where he was tasked with creating new, low-power microprocessors to help it expand beyond chips for personal computers and servers.

Tenstorrent is four-year-old startup that’s trying to develop a processor that’s able to enable a dramatic speed-up in artificial intelligence training and inference, or the process of running trained machine learning models. The company has built a chip called “Grayskull” made up of 120 “Tensix” cores that focus on the mathematical operations common to AI inference, such as matrix multiplications.

Tenstorrent is led by another well-known name in chip design, co-founder and Chief Executive Ljubisa Bajic, who previously worked with Keller at AMD. He also played a part in designing AMD’s Zen architecture and has also worked for Nvidia Corp.

Tenstorrent faces a lot of competition as it bids to create a better chip for AI. The likes of Intel, AMD and Nvidia are all competitors, as are startups such as Cerebras Systems Inc., Graphcore Inc. and Run:AI Inc., and they all have a different approach.

For example, Nvidia has packed a whopping 54 billion transistors onto its Ampere AI chip so it can tackle multiple tasks at once. On the other hand, Cerebras, instead of making transistors smaller, has opted to make physically larger chips that can hold 1.2 trillion transistors each.

Keller will be tasked with guiding Tenstorrent toward its vision of creating a machine learning training chip that surpasses its rival’s designs. Its strategy is to use software to do a better job of allocating computing power on each chip to increase efficiency, rather than just piling on more transistors, and it claims that the Grayskull processor is capable of 368 trillion operations per second.

“There is nobody more capable of executing this vision than Jim Keller, a leader who is equally great at designing computers, cultures and organizations,” Bajic said.

“Tenstorrent is unique in the way it does “conditional execution” to make AI processing more efficient,” said Moor Insights & Strategy analyst Patrick Moorhead. “Keller’s move signifies that the company’s tech is the real deal.”

Constellation Research Inc. analyst Holger Mueller said Tenstorrent deserves praise for managing to recruit such a successful chip designer. “It will be interesting to see if he can adapt to working at a much smaller technology company,” Mueller said.

In the meantime, Keller’s endorsement of Tenstorrent’s approach will likely help the company attract more investment and industry partners. “Tenstorrent has made impressive progress, and with the most promising architecture out there, we are poised to become a next-gen computing giant,” Keller said.

Photo: Intel