AI startup Cerebras celebrated the victory of chips as others tried and failed


A new display at the Computer History Museum in Mountain View, California recognizes Cerebras Systems’ “WSE-2” chip, a second-generation chip, as fulfilling a decades-old quest to make a single chip from an entire silicon wafer.

Computer History Museum

Technology is one of the most conservative practices in the world, meaning that every invention builds on the successes and failures that came before it.

On Wednesday, artificial intelligence startup Cerebras Systems was honored to continue this tradition at a ceremony at the Computer History Museum in Mountain View, California. The museum has put up a screen displaying “Wafer-Scale Engine 2” or WSE-2, the second version of the company’s AI chip that is the largest computer chip ever made. The chip was introduced last year to power new versions of Cerebras’ supercomputer, the CS-2.

Andrew Feldman, co-founder and CEO of Cerebras, said in an interview with ZDNet via Zoom.

“The scale of what you did is very powerful,” Dan Lewin, president and CEO of the Computer History Museum, said in the same interview with Feldman. “This is a milestone in a journey forward, the implications of which are amazing,” Lewin added.

Blog post On the museum’s web page he announced the recognition, Cerebras issued a press release.

also: Artificial intelligence-powered computer maker Cerebras selects TotalEnergies SE as its first energy customer


Cerebras Systems Product Manager for AI Natalia Vassilieva holds the company’s WSE-2, a single chip that measures approximately the entire surface of a 12-inch half-condor chip. First unveiled in April of 2021, the chip is the heart of the CS-2 Machine, the company’s second version of its custom AI computer.

Cerebras . systems

Lewin said Cerebras’ arrival is important, not only for technical achievement but also for its effects on humans, something he emphasizes in his management of the museum.

“The question I ask is, What does it mean to be human in a non-existent world without computing?Lowen said.

also: AI Chip Startup Cerebras Takes $250 Million Series F In Over $4 Billion

Lewin said humanity is at a tipping point where computer technology can either help solve big problems like the climate, or lead to a form of enslavement.

“The whole point, from my accumulated experience in the industry – if I go back to [Douglas] Engelbart [inventor of the computer mouse] The rationale for mother of all offers Well, we humans create problems that affect us greatly.

“We have a set of real problems on a global scale,” Lewin said. He suggested that technology could help or make matters worse.

also: Cerebras is preparing for an era of 120 trillion parameters of neural networks

Daniel Lewin 2022

“Being able to improve computing in this way” has “amazing” implications for humanity, including addressing “a range of real problems on a global scale,” says Dan’l Lewin, president and CEO of the Computer History Museum.

“Being able to improve computing in this way, we hope will teach us that there are more potential positive uses for these technologies than the unfortunate things that have happened as a result of some of the business models that have come to the fore,” he said of Cerebras chip. truly “.

The WSE-2 chip and its predecessor, introduced three years ago, represent a landmark achievement in the history of manufacturing transistors, the building block of all electronic devices, as an integrated part. The first “planar” integrated circuits, transistors fabricated together as a single factory body, a breakthrough made in the early 1960s simultaneously by Texas Instruments engineer Jack Kilby and Intel founder Bob Noyce, putting together only a handful of transistors.

also: Cerebras continues ‘absolute dominance’ of edge computing, it says, with the world’s largest 2-point chip

Then, in 1965, another founder of Intel, Gordon Moore, hypothesized that refined manufacturing methods would lead to an exponential increase in the number of transistors embedded in a single silicon chip. His guess turned out to be correct, and it became known as “Moore’s Law”. The phenomenon of transistor growth has made the digital age possible, from microcomputers to personal computers to smartphones to data networks to electronics embedded in vehicles and the Internet of Things.

WSE-2 chip It uses 2.6 trillion transistorsnearly fifty times the largest GPU chip today from Nvidia, on a 46-square-millimeter silicon substrate, nearly the entirety of the 12-inch semiconductor chip from which many chips are typically cut.

The chip contains 850,000 individual “cores” for processing AI instructions in parallel.

also: Startup Cerebras tells Supercomputing Conference “We can solve this problem in a period of time that no number of GPUs or CPUs can achieve.”

Andrew Feldman 2022

“The discipline has been in part not taking the prevailing wisdom that this can’t be done, and actually looking at what can’t be done and when, and what progress has been made,” says one of the founders of Cerebras, since that time, CEO Andrew Feldman of Cerebras Approach To solve the problem of chip size.

Cerebras . systems

WSE Technology realizes the decades-old pursuit of the chip world, to make a single chip that takes advantage of an entire chip. Cerebras’ success came in part from returning to previous failures and finding a new way to approach the problem.

Cerebras’s Feldman said, “It means so much to us that this institution has recognized the scope of effort that has previously been shattered and burned in other incarnations, even by some of our industry’s founding fathers – even Jane Amdahl couldn’t get it to work.”

Gene Amdahl, the computer pioneer in mainframes, tried and failed in the late 1980s to make such a monoblock. The very general impression formed in the chip industry from his attempt was that making a single chip the size of a chip was too difficult to be practically impossible.

“When we went back and rechecked, part of the discipline was to not take the prevailing wisdom that this can’t be done, and actually look at what can’t be done and when, and what progress has been made,” Feldman said of the Cerebras approach.

also: Glaxo’s research in biology using the new Cerebras machine shows that the devices could change the way AI works

“When you look back and say Jane Amdahl failed at the chip level, he was building it on a two-inch chip,” Feldman explained. “His chips were smaller than anything everyone makes today, nobody thinks about it, and the tools they have.”

Advances since that time in silicon fabrication processes, and chip design software tools, meant that the chip scaling effort was more feasible when Cerebras hit the problem thirty years later.

“The number of items we use that have been invented by others in order to take a big step forward is enormous,” Feldman said.

In terms of implications, Feldman, a serial entrepreneur in the fields of networking and computing technology, also recognizes that the ramifications of their accomplishments are often far from inventors and entrepreneurs.

also: Cerebras teases the second generation of chip-wide AI

“In the early part of my career, we built early switches and routers that made switching IP virtually free, and we all—Cisco, Juniper, 3Com—never thought something like WhatsApp would change the world,” he said, referring to Meta Properties’ free internet communications app. .

“We knew good things would happen if you made the connection almost free,” he said. “When you set out on this path, all sorts of other things advance and build on top of you, and others do things you cannot imagine.”

Lewin said the Cerebras hack has to do with expanding people’s access to digital tools that were previously the captive domain of professionals.

also: Cerebras hasn’t spent a minute working on MLPerf, CEO says

“Industries are going through a transition from vertical to horizontal,” Lewin said. “There were these huge industries that were very vertical that were targeting the automation of rational tasks like the word processing industry, and CAD/CAM as an industry,” he said, referring to computer-aided design. The advancement of microprocessors led to the emergence of horizontal applications such as Microsoft Office that made those previously vertical industries “friendly by a lot of people”.

Lewin said the museum award is as much a peek into the future as it is a rerun of the past. “History is not about the past, it is about the present to have a conversation with the past,” Lewin said.

He gave an example of the creative dialectic between hardware and software.

also: Cerebras’ first customer for the giant AI chip is the US Department of Energy’s supercomputing pig

At an awards ceremony a few years ago, programming pioneer Grady Buch was honored for his many inventions such as the Unified Modeling Language. “It’s programs, programs, programs,” Lewin recalls.

“Gordon Moore got up and smiled, and said, ‘The program is interesting, but it has to be run on something,'” Lewin recalls.

“So this accelerated handshake and opportunity, by taking advantage of this tremendous capacity and improving it in this way, will help drive these changes, these transitions from horizontal to vertical — compressed in time.”

Feldman said WSE-2 technology should be good for a while, as a milestone along the journey. “I think this is the best we have even quantum [computing] It has arrived, and I am very relieved that it will take some time.”

also: The chief investor says Cerebras has at least three years of competition with the AI ​​giant’s chip

Much like receiving a lifetime achievement award, the honor of being in a museum can seem like the end of things. Feldman expressed his conviction that museum recognition is the beginning of things.

Feldman noted that “museums are often repositories of the past.” “We’re now building a great company on the back of some groundbreaking technologies, and don’t forget that: both are tough.”

A discussion between Lewin and Feldman takes place at 2:30 PM Pacific today and can be watched online On the Museum’s YouTube page.

also: Cerebras CEO talks about the big implications of machine learning for the company’s big slice