Cerebras Wafer Scale Engine AI chip can take on Nvidia - Protocol

2022-06-24 16:42:55 By : Ms. wonvi audio

Chip startup Cerebras has developed a foot-wide piece of silicon, compared to average chips measured in millimeters, that makes training AI cheap and easy.

At the core of Cerebras’ pitch is a chip that is roughly the size of a dinner plate.

Inside a conference room at a Silicon Valley data center last week, chip startup Cerebras Systems founder and CEO Andrew Feldman demonstrated how the company’s technology allows people to shift between deploying different versions of an AI natural language model in a matter of moments, a task that usually takes hours or days.

“So we’ve made it 15 keystrokes to move among these largest models that have ever been described on a single machine,” Feldman said.

This, to Feldman and Cerebras, represents a triumph worth noting. Cerebras claims the system that achieved this feat has also accomplished a world first: It can train an entire 20-billion-parameter model on a single nearly foot-wide superchip. Without its technology, the company said scaling an AI model from 1 billion parameters to 20 billion parameters might require users to add more server hardware and reconfigure racks inside of a data center.

Training a natural language AI model on one chip makes it considerably cheaper and delivers a performance boost that is an order of magnitude superior to Nvidia’s flagship graphics processor-based systems, Feldman said. The idea is to give researchers and organizations with tiny budgets — in the tens of thousands of dollars range — access to AI training tools that were previously only available to much larger organizations with lots of money.

“Models have grown really fast in this area. Language processing, and the challenges of delivering compute for these models, is enormous,” Feldman said. “We sort of have made this class of model practical, useful to a whole slice of the economy that couldn’t previously do interesting work.”

The AI models that Feldman is talking about are simply methods of organizing mathematical calculations by breaking them up into steps and then regulating the communication between the steps. The point is to train a model to begin to make accurate predictions, whether that’s the next piece of code that should be written, what constitutes spam and so on.

AI models are typically large to begin with, but those built around language tend to be even larger. For language models, context — as in more text, such as adding an author’s entire body of work to a model that began with a single book — is crucial, but that context can make them far, far more complex to operate. Market-leader Nvidia estimates that AI tasks have spurred a 25-fold increase in the need for processing power every two years.

This exponential increase has led companies like Cerebras and others to chase AI as a potential market. For years, hardware investments were seen as bad bets among venture capitalists who were only willing to fund a few promising ideas. But as it became clear that AI as a class of computation would open the door for fresh ideas beyond the general purpose processors made by the likes of Intel and Nvidia, a new class of startups was born.

Cerebras, which is Latin for “mind,” is one of those startups. Founded in 2015, Feldman and his team, which includes a number of AMD veterans in key technology roles, have raised roughly $735 million — including funding from the CIA venture arm In-Q-Tel, the CEO said — at a $4.1 billion valuation.

At the core of Cerebras’ pitch is a chip that is roughly the size of a dinner plate, or an entire foot-wide silicon wafer, called the Wafer Scale Engine.

The idea of a wafer-size chip like the one that powers Cerebras’ systems isn’t a novel concept; similar ideas have been floating around for decades. A failed bid by Trilogy Systems in the early 1980s that raised roughly $750 million in today’s dollars is one notable attempt at a superchip, and IBM and others have studied the idea but never produced a product.

But together with TSMC, Cerebras has settled on a design that could be fabricated into a functioning wafer-size chip. In some ways, Cerebras is almost two startups stuck together: It’s interested in tackling the growing challenge of AI compute, but it has also achieved the technological feat of producing a useful chip the size of a wafer.

A Cerebras CS-2 system running inside a data center.Photo: Max A. Cherney/Protocol

The current generation of what Cerebras calls the WSE-2 can offer considerable performance improvements over stringing together multiple graphics chips to achieve the computational horsepower to train some of the largest AI models, according to Feldman.

“So it's unusual for a startup to have deep fab expertise, [but] we have profound expertise,” Feldman said. “And we had an idea of how they could, within their permitted flexibility in their flow, fit our innovation.”

The advantage of building a chip of that size is that it allows Cerebras to duplicate the performance of dozens of other server chips — roughly 80 graphics processors, for some large AI models — and squishes them onto a single piece of silicon. Doing so makes them considerably faster, because, in part, data can move faster across a single chip than across a network of dozens of chips.

"[Our] machine is built for one type of work,” Feldman said. “If you want to take the kids to soccer practice, no matter how shitty they are to drive, the minivan is the perfect car. But if you've got your minivan and you try and move two-by-fours and 50-pound sacks of concrete, you realize what a terrible machine it is for that job. [Our chip] is a machine for AI.”

This story was updated to correct the amount of money raised by Trilogy Systems.

Are you keeping up with the latest cloud developments? Get the Enterprise team's newsletter every Monday and Thursday.

Your information will be used in accordance with our Privacy Policy

Thank you for signing up. Please check your inbox to verify your email.

Sorry, something went wrong. Please try again.

A login link has been emailed to you - please check your inbox.

Max A. Cherney is a senior reporter at Protocol covering the semiconductor industry. He has worked for Barron's magazine as a Technology Reporter, and its sister site MarketWatch. He is based in San Francisco.

Apple’s “buy now, pay later” product has a distinctly different distribution strategy that means it doesn’t directly threaten Affirm, Klarna and Afterpay.

Apple Pay Later emerges as a distinctly different product than what Klarna and Affirm offer.

Veronica Irwin (@vronirwin) is a San Francisco-based reporter at Protocol covering fintech. Previously she was at the San Francisco Examiner, covering tech from a hyper-local angle. Before that, her byline was featured in SF Weekly, The Nation, Techworker, Ms. Magazine and The Frisc.

Apple’s entry into the “buy now, pay later” market was one of its worst-kept secrets: Analysts had been predicting the company’s rollout of a pay-later service as early as 2020. The most common read on the move was predictable: Apple was here to smash the competition. The company has a track record of jumping into new sectors late and still managing to come out on top — the iPod came out when there were tons of MP3 players on the market.

But some analysts have a starkly different view. When you look at it under the hood, Apple Pay Later emerges as a distinctly different product than what Klarna and Affirm offer, they say — and one that isn’t much of a market predator.

“This is an opportunity to greatly expand ‘buy now, pay later’ services to people who haven’t used them in the past,” said Ian Rasmussen, who co-leads the North American asset-backed securities ratings group at Fitch Ratings.

That’s for two main reasons. For one, Apple is marketing its pay-later offering to customers in an entirely different place than competitors. Second, the customers Apple is targeting have habits that differ markedly from traditional “buy now, pay later” users.

The most popular pay-later providers have a straightforward primary marketing strategy: partner with retailers that typically feature their logos at checkout, prompting customers to pay over time. Many of these partnerships are exclusive, allowing a company like Klarna, for example, to be the only pay-later provider for customers checking out at Macy’s. The tight integration is meant to boost conversions. Reducing or eliminating “friction” is a “huge factor” in helping sign customers up, said Harry Kohl, a director at Fitch Ratings.

The pay-later companies are trying to extend their reach through financial super apps, which allow customers to apply for payment plans even when there’s not a deal in place with a retailer, though that takes more steps than pressing a button at checkout. Other tools are the virtual and physical cards that allow customers to pay online or in-store, like Affirm’s Debit+ card or the Klarna Card.

Apple Pay Later lives in Apple’s digital wallet, an iPhone feature the company has been trying to build up. Users will be prompted to check out using Apple Pay Later any time they use Apple Pay, or can configure loans directly in the wallet. Customers who do will be able to complete the payment across four charges, spread out over six weeks. The wallet will also show users their Apple Pay Later plans and the payments due over the next 30 days, Apple said.

Confusingly, Apple Pay Later is almost completely separate from Apple’s other venture into consumer credit, the Apple Card. (Apple Card has its own pay-over-time feature, but only for Apple products.) Customers must link payments to a debit card, not a credit card like the Apple Card.

Apple’s advantage is the wide reach of iPhones among consumers and Apple Pay among merchants — particularly at retail, where the pay-later companies are trying to use cards to boost usage.

“It’s a different mindset,” said Patrick DellaValle, director in the financial services practice at Guidehouse. Apple doesn’t need to compete with other pay-later companies for exclusive contracts with retailers, he said, and can focus on continuing to monetize loyal Apple consumers. “They could miss out [on some customers], but it could also be an intentional strategy if you’re going to focus on those customers that will pay upfront with Apple Pay,” he said.

Those existing Apple customers differ from the kind of customer “buy now, pay later” companies have talked about serving: typically a younger person, turned off by credit cards. Third-party surveys consistently show that Apple customers have higher incomes and spend more than Android users.

About three-quarters of pay-later users in the U.S. are Gen Z or millennials, according to a report from eMarketer. Adults who make between $50,000 and $100,000 are most likely to use “buy now, pay later.” Research from The Ascent also shows that 45% of those customers are using “buy now, pay later” to afford something that didn’t already fit into their budgets.

That all points to Apple not making Apple Pay Later to woo existing “buy now, pay later” customers away from the services they use. Instead, it appears the company made a product for the people who love Apple, aimed at persuading them to make more use of the Wallet app and Apple Pay.

Mike Taiano, a senior director at Fitch’s North American Banks Group, sees it simply. “This is another tentacle for them to get into customers’ everyday life,” he said. “I wouldn’t necessarily assume that they automatically go to the top of the leaderboard.”

Veronica Irwin (@vronirwin) is a San Francisco-based reporter at Protocol covering fintech. Previously she was at the San Francisco Examiner, covering tech from a hyper-local angle. Before that, her byline was featured in SF Weekly, The Nation, Techworker, Ms. Magazine and The Frisc.

Nathan Coutinho leads Logitech's global conferencing business strategy and analyst relations. A Swiss company focused on innovation and quality, Logitech designs products and experiences that have an everyday place in people's lives.Coutinho leads strategy and execution of Logitech's video conferencing solutions, from personal solutions to highly-scalable conference rooms.Coutinho has more than 25 years of experience in the IT industry with various roles in executive leadership, consulting, engineering, marketing and technical sales.

Now that most organizations are returning to the office, there are varying extremes – some leaders demand that employees return to the office, with some employees revolting and some rejoicing to be together again. On the other hand, some companies have closed physical offices and made remote work permanent; creating a sigh of relief for some employees and creating frustration for others.

Most of us are somewhere in between, trying our best to take a measured approach at building the right hybrid strategy tailored to company culture. Some seemingly have begun to crack the code, while the majority are grappling with the when, how, why, and who of this new hybrid work reality.

Hybrid work success looks different depending on who you ask. Your company is made up of a cast of players, each with a role critical to a competitive and thriving business, and with an eye on their North Star: employee happiness. How do you appease all those stakeholders so we can all just move on and do our jobs without getting bogged down with the mechanics of it all?

IT: The technology behind hybrid success

Let’s first consider IT, which is the team most bogged down in the mechanics of it all. Heroically having kept workforces running with ad hoc setups during the pandemic, this team is now focused on standardizing technology for the varied mix of remote and in-office employees.

However, employee expectations are completely different than they were pre-pandemic. Technology that was once a nice-to-have is now table stakes not only in conference rooms, but for any space — whether that’s a private office, hot desk, or remote desk.

As a result the IT team is now thinking holistically about how to outfit their hybrid workforces so that each employee has equal access to the highest-quality hardware, software, and solutions. The setups also have to play well in an ecosystem where employees routinely toggle between Zoom, Microsoft Teams, Google Meet, and a number of other web-based platforms for video meetings.

Since the pandemic, facilities teams have redesigned, closed, shrunk, expanded, rebuilt, moved, and retrofitted offices to accommodate the dynamics of an evolving workforce. Whether they’re creating hot desks, huddle rooms, or traditional conference rooms, this team has the difficult task of space-planning for employees who may or may not even come into the physical office on any given day.

Rightsizing, where each meeting space is outfitted for a specific purpose, is top of mind for facilities pros. Reconfiguring rooms to support new hybrid work schedules enables personalization and a safe return to the office. Understanding how employees will use spaces as they come back, and enabling them to easily find and use these spaces will be critical for success.

Human Resources: It’s all about the employee experience

Keeping the Great Resignation at bay is just the start for this team. HR professionals are trying to build and retain a healthy, productive workforce, in all its facets and complications.

Having distributed locations, while sometimes seen as a tremendous benefit, can also bring drawbacks from the lack of face-to-face communication, onboarding, and mentoring. HR leaders are also keenly aware of the lack of equitable experiences some remote employees face when meeting with in-office colleagues.

While some of the HR team’s concerns about employee experiences can be solved with technology (like using easy, intuitive tech to connect teams), the HR team’s challenges stretch into areas that are hard to measure. Cultivating creativity and building unified cultures within hybrid work are massive undertakings with no one-size-fits-all solution.

Finance teams are looking at the not-so-insignificant impact of hybrid work costs. One of the challenges is hybrid work’s coordination problem, as highlighted by Microsoft in its 2022 Work Trends Index Report. Because most organizations have yet to take a data-driven approach to hybrid work, they are continuing to incur the cost of maintaining their campuses while also shelling out work-from-home stipends to a growing virtual workplace.

Our focus at Logitech is helping the industry solve many of these problems.

Making hybrid work successful for everyone requires much more than AI-based innovations. We’ve built solutions that help make the virtual meeting experience more equitable for all, provide analytics and insights into meeting room usage, while bringing simplicity with one touch.

But that’s just the beginning. Listening to all stakeholders — and seeing their challenges through the lens of their role — is the smartest possible start to make this hybrid work actually work. And we’re on it.

Learn more about how Logitech Video Collaboration is making hybrid work for global brands.

Nathan Coutinho leads Logitech's global conferencing business strategy and analyst relations. A Swiss company focused on innovation and quality, Logitech designs products and experiences that have an everyday place in people's lives.Coutinho leads strategy and execution of Logitech's video conferencing solutions, from personal solutions to highly-scalable conference rooms.Coutinho has more than 25 years of experience in the IT industry with various roles in executive leadership, consulting, engineering, marketing and technical sales.

Don’t know what to do this weekend? We’ve got you covered.

Our recommendations for your weekend.

Nick Statt is Protocol's video game reporter. Prior to joining Protocol, he was news editor at The Verge covering the gaming industry, mobile apps and antitrust out of San Francisco, in addition to managing coverage of Silicon Valley tech giants and startups. He now resides in Rochester, New York, home of the garbage plate and, completely coincidentally, the World Video Game Hall of Fame. He can be reached at nstatt@protocol.com.

We’re getting the weekend started early. Two of our favorite shows are back, and we’re digging a breakout hit vampire game that’s being called a “bullet heaven” and is only $3 on Steam.

Indie developer Luca Galante’s Vampire Survivors is one of the most unlikely breakout hits of the year. The influential roguelike shoot-‘em-up has been called a “bullet heaven,”' in contrast to the bullet hell-style manic shooters in which you must dodge a near-endless stream of projectiles. In Vampire Survivors, the projectiles come from you as you maneuver away from small armies of enemies. The game shoots for you, while most of the fun comes from traversing custom maps and unlocking and upgrading unique characters. It’s hard to describe the appeal without trying the game yourself, but at just $3 on Steam, it’s well worth a try. The game was also added to Game Pass for PC last month.

The adaptation of My Chemical Romance frontman Gerard Way’s peculiar superhero graphic novel series returned this week for a third season. After season two’s time travel shenanigans, “The Umbrella Academy” has officially strayed into alternate universe territory, rife with some headache-inducing paradoxes: It’s all getting a bit overwhelming. Thankfully, the third season is grounded by some excellent performances, most notably by Elliot Page, who worked with writer Thomas Page McBee to incorporate his real-world transition into the fictional narrative.

To many casual observers, Axie Infinity looks like a wondrous success story, one of the first play-to-earn games to successfully deploy all the blockchain bells and whistles of Web3 like NFTs, cryptocurrency and virtual land. But the lesser-known story of its downfall over the past six months is a much more important tale, and one told in precise detail by Rest of World’s Darren Loucaides in an excellent feature published this week. The piece, flush with interviews with the company’s founders, tells the story of how Axie Infinity developer Sky Mavis rose to fame as a poster child of the blockchain gaming movement, and the perils of a fledgling industry rife with hacks and scams and intertwined with an uncontrollable and volatile financial market.

The sixth and final season of Steven Knight’s historical crime drama “Peaky Blinders” is here, having aired in its entirety on the BBC and appeared on Netflix earlier this month. Like prior seasons, season six can seem at first glance like a too-quick six episodes, especially given the length of Netflix’s many other series. But “Peaky Blinders” packs extraordinary amounts of depth into each of those hours as it explores the machinations of Thomas Shelby and his once-scrappy and now terrifyingly powerful criminal organization. If you’ve never watched it, now is a good time to dive in before Knight’s planned feature film wraps the series for good.

A version of this story also appeared in today’s Entertainment newsletter; subscribe here.

Nick Statt is Protocol's video game reporter. Prior to joining Protocol, he was news editor at The Verge covering the gaming industry, mobile apps and antitrust out of San Francisco, in addition to managing coverage of Silicon Valley tech giants and startups. He now resides in Rochester, New York, home of the garbage plate and, completely coincidentally, the World Video Game Hall of Fame. He can be reached at nstatt@protocol.com.

He's been extremely active over the past seven days.

Zuckerberg is envisioning a digital commerce empire.

Nick Statt is Protocol's video game reporter. Prior to joining Protocol, he was news editor at The Verge covering the gaming industry, mobile apps and antitrust out of San Francisco, in addition to managing coverage of Silicon Valley tech giants and startups. He now resides in Rochester, New York, home of the garbage plate and, completely coincidentally, the World Video Game Hall of Fame. He can be reached at nstatt@protocol.com.

It’s been a big week for Mark Zuckerberg’s personal Facebook page. The chief executive of Meta has in the past few years transformed his social network account into a press release distribution center, featuring personalized messages about news, product updates and company announcements and usually only ones Zuckerberg himself is pretty excited about. And now hardly a week goes by without some major Facebook post detailing a new product initiative or update about the topic Zuckerberg is most passionate about: the metaverse.

Zuckerberg has been extremely active over the past seven days, posting four major company announcements about metaverse-related news.

The Meta Pay news is a big deal. Before Facebook was Meta, the company tried and failed both publicly and spectacularly to get an ambitious digital currency and crypto platform off the ground. But it failed to woo regulators and eventually shut everything down in January.

Zuckerberg is envisioning a digital commerce empire. If Facebook the product became one of the world’s most effective and lucrative advertising machines, then the metaverse of Zuckerberg’s dreams will be the largest and most dynamic shopping mall ever made.

This isn’t anything we haven’t heard from the Meta CEO over the past nine months, since rebranding the company as Meta. But it does help contextualize many of this week’s announcements. The New York Times reported yesterday that since the metaverse shift, Zuckerberg has been notably less interested in what preoccupied him largely in the aftermath of the 2016 U.S. election, like election integrity, Facebook’s reputational issues and data privacy scandals.

Instead, the company and its most influential decision-maker are now laser-focused on the metaverse. It’s how Zuckerberg imagines Meta will build the next multibillion-user platform, how the company will make most of its money when or if Facebook and even Instagram’s user bases mostly move on to greener pastures and how it will avoid the pitfalls of building its business on platforms, like mobile, that it does not control.

Nick Statt is Protocol's video game reporter. Prior to joining Protocol, he was news editor at The Verge covering the gaming industry, mobile apps and antitrust out of San Francisco, in addition to managing coverage of Silicon Valley tech giants and startups. He now resides in Rochester, New York, home of the garbage plate and, completely coincidentally, the World Video Game Hall of Fame. He can be reached at nstatt@protocol.com.

Heat pumps could keep you and the climate cool.

A confluence of factors has brought heat pumps from the HVAC shadows into the mainstream over the past few years.

Brian ( @blkahn) is Protocol's climate editor. Previously, he was the managing editor and founding senior writer at Earther, Gizmodo's climate site, where he covered everything from the weather to Big Oil's influence on politics. He also reported for Climate Central and the Wall Street Journal. In the even more distant past, he led sleigh rides to visit a herd of 7,000 elk and boat tours on the deepest lake in the U.S.

Everyone from the president to the International Energy Agency to Google users simply cannot stop talking about heat pumps.

A confluence of factors has brought heat pumps from the HVAC shadows into the mainstream over the past few years. But things really came together for heat pumps this month when Biden signed off on using the Defense Production Act to spur more heat pump production. With the prospect of more heat pumps on shelves (or wherever heat pumps are stored), now’s as good a time as ever to understand the climate-protecting technology that just so happens to save people money and keep homes comfortable in all seasons.

At its simplest, a heat pump moves heat from one place to another. The concept is relatively ancient history, dating back to 1852, though it took a while before heat pumps came into existence.

Today, there are two main flavors of heat pumps: air source and ground source. They work basically the same way, pumping or dumping heat from the air or ground. For the ground source version, a series of coils or a long pipe are installed underground and filled with antifreeze. The antifreeze circulates and draws on the ground’s constant temperature in the 50s. In the winter, that antifreeze is pushed through a compressor that turns it into a gas, a process that heats it up, before it’s piped over a fan that sends heat into a home while cooler air is drawn out. The system works in reverse in the summer. Neat!

If you read the above explanation, you may have noticed one thing missing in the description of how heat pumps keep a house cozy in winter: fossil fuels. And that is why heat pumps are the stuff of climate dreams.

The Biden administration wants to reduce U.S. greenhouse gas emissions by at least 50% by 2030. There are many avenues to do that, but decarbonizing buildings is a vital one. Buildings are responsible for 13% of American greenhouse gas pollution. While overall building emissions peaked in the mid-2000s, speeding up the decline is vital to protecting the climate.

Supply chain woes have also come for the heat pump, sadly. But Biden invoking the Defense Production Act could help make them a bigger priority, opening things up a bit.

The up-front cost of a heat pump is also a challenge. Geothermal heat pumps can run into the tens of thousands of dollars depending on the location. Air source ones are cheaper, and mini splits are cheaper still, though you’ll need one for each zone of your home so costs can add up.

That’s not to say investing in a heat pump is a bad idea; an analysis by Carbon Switch found that a heat pump would save the average household $557 per year on the utility bill. That analysis was done before the Russian war in Ukraine caused gas prices to skyrocket, so the savings could be even greater today.

They’re also not super great in really cold climates, though the Department of Energy has a program to try and fix that.

The Defense Production Act alone isn’t going to spur heat pump mass adoption. Some states offer incentives already; New York, for example, has an array of incentives, including thousands of extra dollars in rebates for low-income homes. Getting more states — and utilities — on board with policies that help people prioritize installing heat pumps over their fossil-fueled counterparts is essential.

Brian ( @blkahn) is Protocol's climate editor. Previously, he was the managing editor and founding senior writer at Earther, Gizmodo's climate site, where he covered everything from the weather to Big Oil's influence on politics. He also reported for Climate Central and the Wall Street Journal. In the even more distant past, he led sleigh rides to visit a herd of 7,000 elk and boat tours on the deepest lake in the U.S.

To give you the best possible experience, this site uses cookies. If you continue browsing. you accept our use of cookies. You can review our privacy policy to find out more about the cookies we use.