Connect with us

Mobile

The five technical challenges Cerebras overcame in building the first trillion transistor chip – TechCrunch

Published

on

Superlatives abound at Cerebras, the until-today stealthy next-generation silicon chip company looking to make training a deep learning model as quick as buying toothpaste from Amazon. Launching after almost three years of quiet development, Cerebras introduced its new chip today — and it is a doozy. The “Wafer Scale Engine” is 1.2 trillion transistors (the most ever), 46,225 square millimeters (the largest ever), and includes 18 gigabytes of on-chip memory (the most of any chip on the market today) and 400,000 processing cores (guess the superlative).

Cerebras’ Wafer Scale Engine is larger than a typical Mac keyboard (via Cerebras Systems)

It’s made a big splash here at Stanford University at the Hot Chips conference, one of the silicon industry’s big confabs for product introductions and roadmaps, with various levels of oohs and aahs among attendees. You can read more about the chip from Tiernan Ray at Fortune and read the white paper from Cerebras itself.

Superlatives aside though, the technical challenges that Cerebras had to overcome to reach this milestone I think is the more interesting story here. I sat down with founder and CEO Andrew Feldman this afternoon to discuss what his 173 engineers have been building quietly just down the street here these past few years with $112 million in venture capital funding from Benchmark and others.

Going big means nothing but challenges

First, a quick background on how the chips that power your phones and computers get made. Fabs like TSMC take standard-sized silicon wafers and divide them into individual chips by using light to etch the transistors into the chip. Wafers are circles and chips are squares, and so there is some basic geometry involved in subdividing that circle into a clear array of individual chips.

One big challenge in this lithography process is that errors can creep into the manufacturing process, requiring extensive testing to verify quality and forcing fabs to throw away poorly performing chips. The smaller and more compact the chip, the less likely any individual chip will be inoperative, and the higher the yield for the fab. Higher yield equals higher profits.

Cerebras throws out the idea of etching a bunch of individual chips onto a single wafer in lieu of just using the whole wafer itself as one gigantic chip. That allows all of those individual cores to connect with one another directly — vastly speeding up the critical feedback loops used in deep learning algorithms — but comes at the cost of huge manufacturing and design challenges to create and manage these chips.

CS Wafer Sean

Cerebras’ technical architecture and design was led by co-founder Sean Lie. Feldman and Lie worked together on a previous startup called SeaMicro, which sold to AMD in 2012 for $334 million. (Via Cerebras Systems)

The first challenge the team ran into according to Feldman was handling communication across the “scribe lines.” While Cerebras chip encompasses a full wafer, today’s lithography equipment still has to act like there are individual chips being etched into the silicon wafer. So the company had to invent new techniques to allow each of those individual chips to communicate with each other across the whole wafer. Working with TSMC, they not only invented new channels for communication, but also had to write new software to handle chips with trillion plus transistors.

The second challenge was yield. With a chip covering an entire silicon wafer, a single imperfection in the etching of that wafer could render the entire chip inoperative. This has been the block for decades on whole wafer technology: due to the laws of physics, it is essentially impossible to etch a trillion transistors with perfect accuracy repeatedly.

Cerebras approached the problem using redundancy by adding extra cores throughout the chip that would be used as backup in the event that an error appeared in that core’s neighborhood on the wafer. “You have to hold only 1%, 1.5% of these guys aside,” Feldman explained to me. Leaving extra cores allows the chip to essentially self-heal, routing around the lithography error and making a whole wafer silicon chip viable.

Entering uncharted territory in chip design

Those first two challenges — communicating across the scribe lines between chips and handling yield — have flummoxed chip designers studying whole wafer chips for decades. But they were known problems, and Feldman said that they were actually easier to solve that expected by re-approaching them using modern tools.

He likens the challenge though to climbing Mount Everest. “It’s like the first set of guys failed to climb Mount Everest, they said, ‘Shit, that first part is really hard.’ And then the next set came along and said ‘That shit was nothing. That last hundred yards, that’s a problem.’”

And indeed, the toughest challenges according to Feldman for Cerebras were the next three, since no other chip designer had gotten past the scribe line communication and yield challenges to actually find what happened next.

The third challenge Cerebras confronted was handling thermal expansion. Chips get extremely hot in operation, but different materials expand at different rates. That means the connectors tethering a chip to its motherboard also need to thermally expand at precisely the same rate lest cracks develop between the two.

Feldman said that “How do you get a connector that can withstand [that]? Nobody had ever done that before, [and so] we had to invent a material. So we have PhDs in material science, [and] we had to invent a material that could absorb some of that difference.”

Once a chip is manufactured, it needs to be tested and packaged for shipment to original equipment manufacturers (OEMs) who add the chips into the products used by end customers (whether data centers or consumer laptops). There is a challenge though: absolutely nothing on the market is designed to handle a whole-wafer chip.

CS Wafer Inspection

Cerebras designed its own testing and packaging system to handle its chip (Via Cerebras Systems)

“How on earth do you package it? Well, the answer is you invent a lot of shit. That is the truth. Nobody had a printed circuit board this size. Nobody had connectors. Nobody had a cold plate. Nobody had tools. Nobody had tools to align them. Nobody had tools to handle them. Nobody had any software to test,” Feldman explained. “And so we have designed this whole manufacturing flow, because nobody has ever done it.” Cerebras’ technology is much more than just the chip it sells — it also includes all of the associated machinery required to actually manufacture and package those chips.

Finally, all that processing power in one chip requires immense power and cooling. Cerebras’ chip uses 15 kilowatts of power to operate — a prodigious amount of power for an individual chip, although relatively comparable to a modern-sized AI cluster. All that power also needs to be cooled, and Cerebras had to design a new way to deliver both for such a large chip.

It essentially approached the problem by turning the chip on its side, in what Feldman called “using the Z-dimension.” The idea was that rather than trying to move power and cooling horizontally across the chip as is traditional, power and cooling are delivered vertically at all points across the chip, ensuring even and consistent access to both.

And so, those were the next three challenges — thermal expansion, packaging, and power/cooling — that the company has worked around-the-clock to deliver these past few years.

From theory to reality

Cerebras has a demo chip (I saw one, and yes, it is roughly the size of my head), and it has started to deliver prototypes to customers according to reports. The big challenge though as with all new chips is scaling production to meet customer demand.

For Cerebras, the situation is a bit unusual. Since it places so much computing power on one wafer, customers don’t necessarily need to buy dozens or hundreds of chips and stitch them together to create a compute cluster. Instead, they may only need a handful of Cerebras chips for their deep-learning needs. The company’s next major phase is to reach scale and ensure a steady delivery of its chips, which it packages as a whole system “appliance” that also includes its proprietary cooling technology.

Expect to hear more details of Cerebras technology in the coming months, particularly as the fight over the future of deep learning processing workflows continues to heat up.



Source link

Continue Reading
Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *

Mobile

Hackers release a new jailbreak tool for almost every iPhone – TechCrunch

Published

on

An iPhone hacking team has released a new jailbreak tool for almost every iPhone, including the most recent models, by using the same vulnerability that Apple last month said was under active attack by hackers.

The Unc0ver team released its latest jailbreak this weekend, and says it works on iOS 11 (iPhone 5s and later) to iOS 14.3, which Apple released in December.

Jailbreaking is a cat-and-mouse game between security researchers who want greater control and customizations over their phones, and Apple, which says it locks down iPhones for security. Hackers build jailbreak tools by finding and exploiting vulnerabilities that can lift some of the restrictions that Apple puts in place, like installing apps outside of its app store, which most Android users are already used to.

In a tweet, the jailbreak group said it used its “own exploit” for CVE-2021-1782, a kernel vulnerability that Apple said was one of three flaws that “may have been actively exploited” by hackers. By targeting the kernel, the hackers are able to get deep hooks into the underlying operating system.

Apple fixed the vulnerability in iOS 14.4, released last month, which also prevents the jailbreak from working on later versions. It was a rare admission that the iPhone was under active attack by hackers, but the company declined to say who the hackers were and who they were targeting. Apple also granted anonymity to the researcher who submitted the bug.

The group’s last jailbreak, which supported iPhones running iOS 11 to iOS 13.5, was fixed in a matter of days last year. Apple works quickly to understand and fix the vulnerabilities found by jailbreak groups, since these same vulnerabilities can be exploited maliciously.

Security experts generally advise iPhone users against jailbreaking because it makes the device more vulnerable to attacks. And while keeping your phone up to date may introduce security fixes that remove the jailbreak, it’s one of the best ways of keeping your device secure.


Early Stage is the premiere “how-to” event for startup entrepreneurs and investors. You’ll hear firsthand how some of the most successful founders and VCs build their businesses, raise money and manage their portfolios. We’ll cover every aspect of company-building: Fundraising, recruiting, sales, legal, PR, marketing and brand building. Each session also has audience participation built-in — there’s ample time included in each for audience questions and discussion.

Continue Reading

Mobile

Cappuccino lets you share short, intimate audio stories with your friends – TechCrunch

Published

on

You might call Cappuccino the anti-Clubhouse, but the company has been iterating on its app concept for a couple of years — its CEO doesn’t have any strong opinions on Clubhouse. And it’s true that Cappuccino is an interesting social app on its own. It has been attracting a loyal user base, especially after a TikTok video went viral.

The startup says it is building an app that helps you record podcasts with friends. Many people have discovered podcasts over the past few years. Podcasts let you subscribe to audio shows and listen to episodes on demand.

At first, people subscribe to podcasts because of their interests. But if you talk about podcasts with your friends, they’ll tell you that they like a show in particular because of the personalities of the hosts.

Listening to a podcast is a content consumption experience that feels like nothing else out there. You might watch all videos released by a particular YouTuber and you might think you know a lot about someone’s personal life by following them on Instagram.

But listening to someone for hours at a time with earbuds in your ears is a very intimate experience. When a podcast works, it feels like you’re sitting in a room with a few friends and just listening to what they have to say.

And yet, chances are your favorite podcast hosts are not your friends.

This is where Cappuccino fits. The app lets you create groups with your friends or your families. Members of the group can record a short audio message — a bean, as the startup calls it. They talk about what’s on their mind for a couple of minutes. The next morning, group members receive a notification saying that your morning cappuccino has been brewed.

When you hit play, a chill intro music starts playing followed by audio messages from your friends. It isn’t just a succession of voice memos — it feels like a relaxing mix of happy, funny, caring, thoughtful messages from your friends.

While Cappuccino is a social app, it is focused on your close friends and your family. You aren’t trying to get more followers and you are not sharing public posts. Everything is private by design and focused on groups of real-life friends.

In many ways, it reminds me of Snapchat’s group stories. But Snapchat wasn’t the main inspiration for Cappuccino — it was podcasting.

Image Credits: Cappuccino

Prototype early, iterate often

I talked with the company’s co-founder and CEO Gilles Poupardin about the origin story of the app. Cappuccino isn’t Poupardin’s first startup. He had worked on Whyd for several years and lived the full startup experience — he raised founding rounds, chose to pivot, attended Y Combinator in San Francisco, parted ways with his company’s CTO and chose to shutter the startup.

Among other things, Whyd worked on a voice-controlled connected speaker before Amazon’s Echo product lineup and Google’s Nest speakers really took off. It’s hard to compete with tech giants, even harder when you’re competing on the hardware front.

After that, the Whyd team worked on a service that lets you create your own voice assistant. That didn’t really take off as expected either.

During the summer of 2019, Olivier Desmoulin reached out to Poupardin. Back then, Desmoulin was heading design for Jumbo, an app that helps you stay on top of your online privacy.

“At the time, I didn’t know if I wanted to start a company again — I pivoted 15 times [with Whyd],” Poupardin told me.

But they started discussing about podcasts and AirPods — and audio at large — as the next frontier for social apps. The basic premise was simple. A lot of people were listening to podcasts, but very few people were creating their own podcasts.

There are three reasons why your neighbor doesn’t have its own podcast but sometimes posts stuff on Instagram and Snapchat:

  • Podcasts are long-form content
  • It’s technically complicated to record and release a podcast
  • You are trying to attract an audience of people who don’t know you.

With Cappuccino, the idea is to take a reverse stance on these three points: short content, easy to record and personal. It’s supposed to be a better experience for both people recording audio and people listening to audio.

The first version of Cappuccino isn’t an app, it’s a side project. “We created a group on WhatsApp, we invited 10 to 15 people and we asked them to record voice memos and send them all to Olivier,” Poupardin told me.

Every night, Olivier Desmoulin would fire up GarageBand and create a mix of all voice memos. In the morning, he would send a message to the group conversation on WhatsApp and write: “Hey, your cappuccino is here.”

Image Credits: Cappuccino

After getting some positive feedback from group members, Pouparding and Desmoulin chose to move forward and create something that feels more like an app. But they both knew that creating a social app was incredibly hard when it comes to attracting users. They developed something quickly so that they weren’t wasting time developing something that nobody would use.

“We built the first version of the app in four days by using a hack — we were using Airtable as the backend service,” Poupardin said.

Once again, feedback from beta users was pretty good. They showed the app to some investors and ended up raising $1.2 million from Alexia Bonatsos (Dream Machine, also a former TechCrunch editor), SV Angel, Kevin Carter (Night Capital), Niv Shrug Capital, Jean de La Rochebrochard (Kima Ventures), Kevin Kuipers, Willy Braun, Marie Ekeland, Solomon Hykes (founder of Docker), Pierre Valade (founder of Sunrise and Jumbo Privacy), Moshe Lifschitz (Basement Fund), Anthony Marnell, Bryan Kim and a bunch of others.

Gawen Arab who was the CTO at Whyd teamed up once again with Poupardin, proving that time is a flat circle. He’s now co-founder and CTO at Cappuccino.

Image Credits: Cappuccino

Letting people talk about you

The Cappuccino team hasn’t been active when it comes to press relations or ads. It’s been a slow build up with some interesting spikes.

Last summer, Product Hunt super user Chris Messina created a post about Cappuccino. It was a bit of a surprise as the startup wasn’t trying to get featured on Product Hunt. Still, the co-founders diligently answered questions from the Product Hunt community.

The following day, Product Hunt’s newsletter featured Cappuccino. It was titled “The next big audio social network?” That brought some new users to the app.

Image Credits: Cappuccino

But things really started to take off when Brittany Kay Collier shared a video on TikTok about Cappuccino a few weeks ago. She sent a direct message to Poupardin on Instagram, telling him that it was attracting a lot of views. The video ended up attracting around 3.8 million views and 850,000 likes.

Two days later, Poupardin sent her a job offer to join the team. He was secretly hoping she would say yes, and she was secretly dreaming about getting a job at a company like Cappuccino.

Over the past couple of weeks, Cappuccino attracted 225,000 new users. They created 130,000 groups and sent around one million audio stories.

When the team is reading public posts about Cappuccino on Twitter, it feels like the app has found its core user base. The most loyal users seem to be young women in their twenties. They want to keep in touch with long-distance best friends.

They might be graduating from college and moving to a different part of the country. They might be stuck at home because of the current pandemic.

And it seems like new users have no issue hitting the record button and telling stories — everybody is familiar with voice messages on WhatsApp and iMessage after all.

“Something that is interesting with audio messages as a medium is that you tell different stories from what you would tell by taking a photo for Instagram, sending a Snap or creating a video on TikTok,” Poupardin said.

But what about the elephant in the room then? Clubhouse has topped 8 million downloads already. Poupardin listed all the differences in social graph, audio format and user base. According to him, there’s enough room for multiple audio apps.

“With video, you have YouTube, Twitch and TikTok — those are all different formats. Audio is potentially going to follow the same trend,” Poupardin said. Social apps first took advantage of the camera in your smartphone, because the camera was the killer hardware feature. And audio seems like the natural next step.

He feels like he isn’t competing with other audio startups for now. He wants people to wake up and listen to Cappuccino instead of random music on Spotify. “It’s going to help people who feel lonely,” he said.

Continue Reading

Mobile

Instagram launches ‘Live Rooms’ for live broadcasts with up to four creators – TechCrunch

Published

on

Instagram today announced it’s adding a much-requested feature to its app with the launch of “Live Rooms,” which allow up to four people to broadcast live together at the same time. Previously, the app only allowed users to live stream with one other person, similar to Facebook Live. The company says it hopes Live Rooms will open up more creative opportunities in terms of live broadcast formats to allow for things like live talk shows, expanded Q&A’s or interviews, jam sessions for musicians, live shopping experiences, and more.

In addition to the ability to live stream with more people, Instagram also touts how the new feature can help creators to make more money. Last year, in the early days of the COVID-19 crisis, Instagram introduced badges as a way for fans to support their favorite creators during a live video. Once purchased, the badges appear next to a fan’s name throughout the live video, helping them to stand out in the comments and unlock other special features, like placement on the creator’s list of badge holders and access to a special heart.

Badges became more broadly available last fall, at three price points: $0.99, $1.99, or $4.99.

With Live Rooms, fans can buy badges to support the hosts (one badge per person) as well as use other interactive features like Shopping and Live Fundraisers. The company says it’s also now developing other tools, like moderator controls and audio features that will roll out in the months to come.

To start a Live Room, you’ll swipe left and select the Live camera option, then title the Room and tap the Room icon to add guests. Here, you’ll see a list of people who’ve already requested to go live with you and you’ll be able to search for other guests to add.

Image Credits: Instagram

When you start the Live Room, you’ll remain at the top of the screen while guests are added. The guests can be added all at once or individually, depending on your preference. This allows for opportunities to add “surprise guests” to live streams to keep fans engaged.

The ability to add more guests to a live stream can also help a creator grow their follower base, as all the guests’ followers are notified about the Live Room, in addition to your own.

For safety reasons, any person that’s been blocked by any of the Live Room participants will not have access to join the live stream. Plus, any guests who have previously had their live access revoked due to violations of Instagram’s Community Guidelines won’t be able to join any Live Rooms.

During live broadcasts, the hosts can also report and block comments and use comment filters to maintain a safer experience for all viewers.

Live broadcasts became an increasingly important way for creators, business owners and brands to stay connected with followers during the pandemic, which shut down in-person live events, including concerts, shows, classes, conferences, meetups, and more. Instagram reported a 70% increase in Live views from February to March, for instance, as creators and businesses shifted their work online.

Image Credits: Instagram

As the pandemic wore on throughout 2020 and into 2021, the lack of in-person connection has allowed for other opportunities and even new social networks to grow. Live audio platform Clubhouse, for example, has seen rapid adoption, particularly by the tech and creative crowds, who today use the app to tune into live shows, chat sessions, and even big-name interviews. Twitter is now building a rival, and reportedly, so is Facebook.

But while Clubhouse offers a very different experience, it still operates in the same broader space of allowing fans to connect with high-profile individuals of some sort — entrepreneurs and founders, celebrities, market experts, thought leaders, influencers, and so on. And because users’ time is limited, seeing this type of activity shift to non-Facebook owned platforms is likely of concern to Instagram and its parent.

Meanwhile, in the live video broadcasting space, Instagram today faces a number of competitors, from those focused on a particular niche — like game streaming site Twitch, live shopping apps, and more— as well as general purpose live platforms offered by YouTube and TikTok. (The latter was spotted offering a four-up live stream format just last month, in fact.)

Instagram says Live Rooms are rolling out now to both iOS and Android to all global markets. The company expects the rollout to reach 100% of its user base within the week.

 

Continue Reading

Trending