In a sign of the profound changes being wrought in computing by artificial intelligence, Toronto-based AI chip startup Tenstorrent on Wednesday announced it has hired legendary chip designer Jim Keller to be its chief technology officer.
Keller most recently served at Intel and before that re-invented the microprocessor architecture at Advanced Micro Devices.
Keller said in prepared remarks, “Software 2.0 is the largest opportunity for computing innovation in a long time. Victory requires a comprehensive re-thinking of compute and low level software.”
Added Keller, “Tenstorrent has made impressive progress, and with the most promising architecture out there, we are poised to become a next gen computing giant.”
Tenstorrent’s co-founder and CEO, Ljubisa Bajic, said, “I am thrilled to be working with Jim and beyond excited about the possibilities our partnership unlocks.”
“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 commented, referring to the company’s quest to achieve what it calls “a transformation in computational capabilities.”
The new workloads of AI, especially deep learning, have the potential to change the entire computing landscape. Startups such as SambaNova Systems, Cerebras Systems, and Graphcore have built entire computing platforms to train, out of a belief that deep learning re-writes fundamental rules of computing.
Four-year-old Tenstorrent has received $41.7 million to develop a chip to dramatically speed up the inference task for AI training. The company’s “Grayskull” chip features 120 of what it calls “Tensix” cores that focus on the mathematical operations common to deep learning forms of AI, such as matrix multiplications.
Also: Startup Tenstorrent shows AI is changing computing and vice versa
Tenstorrent is in the thick of a vibrant market for deep learning inference acceleration, including Nvidia’s A100 chips, of course, but also Intel’s Habana product, and a gaggle of startup companies that keeps expanding, such as SimpleMachines and Groq.
Early results of the Grayskull part garnered some favorable commentary last spring. In a review of the Grayskull part, the Linley Group’s lead analyst, Linley Gwennap, wrote that the Grayskull part has “excellent” performance relative to Nvidia and other startups, including Groq.
Tenstorrent’s “Grayskull” chip uses 120 cores to compute the matrix multiplication that define the bulk of deep learning workloads.
Tenstorrent.Keller resigned from Intel last June for “personal reasons,” the chip giant announced, with a six-month stint to follow as an advisor to the company.
Before coming to Intel in 2018, Keller worked at Tesla helping the car maker design custom chips for self-driving systems. But he is better known for many landmark projects in processor design, including the Zen architecture that underlies AMD’s server and PC processors; and Apple’s first custom processor for its iOS devices was built under the direction of Keller after Apple acquired Keller’s chip startup, PA Semiconductor, in 2008.
Also: ‘It’s not just AI, this is a change in the entire computing industry,’ says SambaNova CEO
Analyst Gwennap speculated that Keller came to Intel to work on the code-named Sapphire Rapids, which was to be Intel’s “first from-scratch x86 microarchitecture in more than a decade,” as Gwennap wrote back then.
Keller’s departure from Intel was greeted with shock and dismay last summer. As Rosenblatt Securities chip analyst Hans Mosesmann wrote at the time, “Keller’s departure is a big deal and suggests that whatever he was implementing at Intel was not working or the old Intel guard did not want to implement it.”
