(Bloomberg) — Cerebras Systems Inc. startups to build the first nine AI supercomputers with money from Abu Dhabi, as part of an effort to provide alternatives to systems using Nvidia Corp. technology.
The Condor Galaxy 1, located in Santa Clara, Calif., is now operational, according to Cerebras founder and CEO Andrew Feldman. He said the supercomputer, which cost more than $100 million, will double in size “in the coming weeks.” New systems will follow in Austin and Asheville, North Carolina in the first half of next year, with outdoor locations coming online in the second half of 2024.
The project is part of a push to add computing power to artificial intelligence services, which require the kind of processing intensive that has become a specialty of Nvidia, the world’s most valuable chipmaker. The Cerebras machine, which Feldman describes as the largest purpose-built AI computing center, is an attempt to address this need with a new approach.
It also marks a deeper push into the field by the United Arab Emirates, which is banking on next-generation technology with a company called Group 42, or G42. The company is focused on driving synthetic research into practical uses in areas such as aerospace and healthcare.
The new supercomputers will be powered by Cerebras and used in G42 projects. Any excess capacity will be offered commercially as a service.
For Cerebras, which is based in Silicon Valley, the new systems provide exposure that it hopes will lead to wider adoption. The company’s offering is based on massive chips made of all-silicon wafers — disks that are typically sliced to create multiple components.
Read more: Artificial intelligence startup raises $4 billion
Feldman argues that his processors have the advantage of being able to handle large sets of data at once, rather than only working on bits of information at a time. And compared to Nvidia processors, it also requires less complex software needed to run the chips in concert.
This year, cloud computing providers such as Microsoft Corp. and Amazon.com’s AWS is stocking Nvidia processors to meet the rampant demand for OpenAI’s ChatGPT and other AI tools. Nvidia has about 80% of the market for so-called accelerators that help manage these workloads.
With his take on computing, Feldman aims to prove that the explosion of AI won’t just benefit giant tech companies that can afford big-budget gear.
“There’s a misconception that there are only seven to ten companies in the world that can buy at scale to make a difference,” he said. “It radically changes the conversation. »
Feldman processors are too big to fit in conventional machines, prompting Cerebras to introduce their technology into custom-built computers. The machines are also based on industry-standard processors from Advanced Micro Devices Inc. — the company that bought Feldman’s former startup SeaMicro Inc.
One of the new supercomputers will be able to train programs on data sets of 600 billion variables, Cerebras said, with the ability to scale that up to 100 trillion. Each will be made up of 54 million compute cores optimized for artificial intelligence.
© 2023 Bloomberg LP
