Video yükleniyor...

Video Yüklenemedi

Ana Sayfaya Dön

CS4 uses the same wafer as CS3. So where does the extra speed come from? You asked on Twitter and Discord. Join Angela Yeung, SVP of product and Sarah Chieng, on the behind-the-scenes on the engineering behind CS4: connecting wafers, recirculating cooling water, and getting compute deployed at scale....

cerebras's profile picture

Cerebras

78,225 subscribers

30,508 görüntüleme • 1 ay önce •via X (Twitter)

Bilim & Teknoloji
LIVE

Anya Rossi• Live Now

Private livecam show

2 Yorum

Steve profil fotoğrafı
Steve1 ay önce

@MilksandMatcha As a shareholder love the content

FrencHorizon · Infra IA profil fotoğrafı
FrencHorizon · Infra IA1 ay önce

@MilksandMatcha @grok resume la vidéo

Benzer Videolar

Connecting the dies and working around defects were fundamental challenges for wafer-scale computing. Making it work also came down to power, cooling, and reliability. Sean Lie explains how those lessons are shaping Cerebras’ approach to stacking DRAM. From Sean Lie’s conversation with swyx on Latent.Space.
1:01

Connecting the dies and working around defects were fundamental challenges for wafer-scale computing. Making it work also came down to power, cooling, and reliability. Sean Lie explains how those lessons are shaping Cerebras’ approach to stacking DRAM. From Sean Lie’s conversation with swyx on Latent.Space.

Cerebras

12,407 görüntüleme • 1 ay önce

We get a lot of questions about CS-4’s components. Take a closer look at the engineering behind the CS-4 Wafer-Scale Backpack. ⤵️
1:10

We get a lot of questions about CS-4’s components. Take a closer look at the engineering behind the CS-4 Wafer-Scale Backpack. ⤵️

Cerebras

300,948 görüntüleme • 27 gün önce

Behind the scenes on THE LORD OF THE RINGS trilogy and the incredbile use of scale models.
0:38

Behind the scenes on THE LORD OF THE RINGS trilogy and the incredbile use of scale models.

All The Right Movies

30,349 görüntüleme • 2 ay önce

When we started Cerebras, we knew we would have to solve difficult problems in yield, power, and cooling. What became clear was that making wafer-scale work would require rethinking almost the entire system. Today, we answered questions from the community about those engineering decisions, how we run larger models, and what it takes to program the hardware. 2:15 Why hasn't NVIDIA gone wafer-scale? 4:10 How we build a working chip out of an imperfect wafer 8:07 What happens when a model won't fit on one wafer? 11:34 Why we chose SRAM over HBM 14:26 How much of a barrier is NVIDIA's CUDA moat?
18:44

When we started Cerebras, we knew we would have to solve difficult problems in yield, power, and cooling. What became clear was that making wafer-scale work would require rethinking almost the entire system. Today, we answered questions from the community about those engineering decisions, how we run larger models, and what it takes to program the hardware. 2:15 Why hasn't NVIDIA gone wafer-scale? 4:10 How we build a working chip out of an imperfect wafer 8:07 What happens when a model won't fit on one wafer? 11:34 Why we chose SRAM over HBM 14:26 How much of a barrier is NVIDIA's CUDA moat?

Sean Lie

33,146 görüntüleme • 1 ay önce

ar0 cs4 hd streams on the menu #SOON 👀
0:32

ar0 cs4 hd streams on the menu #SOON 👀

plasma

12,118 görüntüleme • 7 ay önce

When healthy, where does Acuna rank on the best players in Major League Baseball list? You can argue first based on basepath speed and arm in outfield. I’ve got him third behind Ohtani and Judge. You guys?
0:36

When healthy, where does Acuna rank on the best players in Major League Baseball list? You can argue first based on basepath speed and arm in outfield. I’ve got him third behind Ohtani and Judge. You guys?

Clay Travis

230,979 görüntüleme • 1 yıl önce

46,225 mm² of silicon 4 trillion transistors 900,000 AI-optimized cores 250 petaflops of AI compute At the center of every CS-4 Wafer-Scale Backpack is the WSE-3 Turbo ⤵️
0:51

46,225 mm² of silicon 4 trillion transistors 900,000 AI-optimized cores 250 petaflops of AI compute At the center of every CS-4 Wafer-Scale Backpack is the WSE-3 Turbo ⤵️

Cerebras

90,618 görüntüleme • 13 gün önce

On a scale from the Chrysler Building to the Morgan Library, where does the drag king and author of ‘Showbiz!’ Murray Hill fit?
2:49

On a scale from the Chrysler Building to the Morgan Library, where does the drag king and author of ‘Showbiz!’ Murray Hill fit?

New York Magazine

17,712 görüntüleme • 17 gün önce

What is $USDN and where does the yield come from?
2:02

What is $USDN and where does the yield come from?

Everything.inc

11,333 görüntüleme • 1 yıl önce

Six continents. One performer. Music connects cultures; the E2 does the same, connecting communities on a global scale. See us at Farnborough Airshow 2026!
1:00

Six continents. One performer. Music connects cultures; the E2 does the same, connecting communities on a global scale. See us at Farnborough Airshow 2026!

Embraer

1,380,313 görüntüleme • 3 ay önce

AI agents rely on a good harness. Faster inference gives a harness room to inspect results and recover sooner, while you steer the work. Sarah Chieng breaks down the loop.
1:17

AI agents rely on a good harness. Faster inference gives a harness room to inspect results and recover sooner, while you steer the work. Sarah Chieng breaks down the loop.

Cerebras

137,511 görüntüleme • 19 gün önce

500% NDR at Anthropic is the stat of the year for me… I asked Krishna what questions he'd ask the model companies if he were an investor. His three: 1) What is the ROI on compute? 2) How are customers seeing ROI? Are they deploying at scale or just testing? 3) How do you think about compute in the future, and where does it come from? His answer to the second one: "Our net dollar retention is over 500% on an annualized basis. Nine of the Fortune 10 are customers. [In the 20 minute Uber to this podcast], I signed two double-digit million-dollar commits."
2:50

500% NDR at Anthropic is the stat of the year for me… I asked Krishna what questions he'd ask the model companies if he were an investor. His three: 1) What is the ROI on compute? 2) How are customers seeing ROI? Are they deploying at scale or just testing? 3) How do you think about compute in the future, and where does it come from? His answer to the second one: "Our net dollar retention is over 500% on an annualized basis. Nine of the Fortune 10 are customers. [In the 20 minute Uber to this podcast], I signed two double-digit million-dollar commits."

Patrick OShaughnessy

184,390 görüntüleme • 5 ay önce

Where does all the water come from 🤔
2:59

Where does all the water come from 🤔

𝕵𝖔𝖍𝖆𝖓𝖓𝖊𝖘 𝕸𝖎𝖉𝖌𝖆𝖗𝖉𝖗

40,094 görüntüleme • 1 yıl önce

I joined AMD right after graduating from MIT. Two stints there and a decade building Cerebras later, I still lose sleep over how much memory and compute to put on a chip. I’ve spent my entire career in computer architecture, and I’ve never seen the problems change this quickly. Models keep getting larger. The demands on the hardware keep growing. Decisions about where to put memory and how to connect it have become central to what people can build with AI. We chose SRAM for its bandwidth. Now we’re working on wafer-scale stacked DRAM to give larger models more capacity while preserving fast inference. That brings us back to some very physical problems: how to power it, cool it, and manufacture it reliably. This is the work I love. In part two of our AMA, I answered your questions about those decisions and where we’re going next. 0:30 What changes between CS4, CS5, and CS6? 2:19 HBM already stacks memory. What’s different here? 3:34 We chose SRAM for speed. Why add DRAM? 4:52 Drawing the stack is easy. Making it work isn’t. 7:23 The memory-versus-compute decision that keeps me up at night
9:28

I joined AMD right after graduating from MIT. Two stints there and a decade building Cerebras later, I still lose sleep over how much memory and compute to put on a chip. I’ve spent my entire career in computer architecture, and I’ve never seen the problems change this quickly. Models keep getting larger. The demands on the hardware keep growing. Decisions about where to put memory and how to connect it have become central to what people can build with AI. We chose SRAM for its bandwidth. Now we’re working on wafer-scale stacked DRAM to give larger models more capacity while preserving fast inference. That brings us back to some very physical problems: how to power it, cool it, and manufacture it reliably. This is the work I love. In part two of our AMA, I answered your questions about those decisions and where we’re going next. 0:30 What changes between CS4, CS5, and CS6? 2:19 HBM already stacks memory. What’s different here? 3:34 We chose SRAM for speed. Why add DRAM? 4:52 Drawing the stack is easy. Making it work isn’t. 7:23 The memory-versus-compute decision that keeps me up at night

Sean Lie

33,118 görüntüleme • 1 ay önce

On a scale from Fifth Avenue to “plead the Fifth,” where does Carole Radziwill fit?
2:57

On a scale from Fifth Avenue to “plead the Fifth,” where does Carole Radziwill fit?

New York Magazine

165,155 görüntüleme • 1 ay önce

We asked James Wang why Cerebras can run large models so quickly. His answer: Inference speed is mostly a memory problem. Instead of constantly pulling weights from external memory, Cerebras splits the model across multiple wafers and pipelines the layers together. “Inference is all about memory bandwidth.” “You don’t want to store weights external to the wafer, because the second it’s outside, it becomes much slower.” “We put the weights on the inside.” “The team built a new software stack that basically lets us split the models by layer and store them layer by layer across multiple wafers.” “That allows us to never have to read from external memory, and that’s what makes it so fast.” “For context, Claude is about 100 tokens per second.”
1:31

We asked James Wang why Cerebras can run large models so quickly. His answer: Inference speed is mostly a memory problem. Instead of constantly pulling weights from external memory, Cerebras splits the model across multiple wafers and pipelines the layers together. “Inference is all about memory bandwidth.” “You don’t want to store weights external to the wafer, because the second it’s outside, it becomes much slower.” “We put the weights on the inside.” “The team built a new software stack that basically lets us split the models by layer and store them layer by layer across multiple wafers.” “That allows us to never have to read from external memory, and that’s what makes it so fast.” “For context, Claude is about 100 tokens per second.”

MTS

27,935 görüntüleme • 5 ay önce

We gave two agents the same task: “Find images matching this description.” Both use Gemma 4 31B. One runs on Cerebras. The other runs on GPUs. You can see the difference. Speed changes the product experience. What would you build if you didn't have to wait?
1:11

We gave two agents the same task: “Find images matching this description.” Both use Gemma 4 31B. One runs on Cerebras. The other runs on GPUs. You can see the difference. Speed changes the product experience. What would you build if you didn't have to wait?

Cerebras

142,645 görüntüleme • 4 ay önce

6/And, finally, Majorana 2. Majorana 1 proved out the physics of the topological approach, but Majorana 2 is where we start engineering at scale.
1:37

6/And, finally, Majorana 2. Majorana 1 proved out the physics of the topological approach, but Majorana 2 is where we start engineering at scale.

Satya Nadella

38,085 görüntüleme • 5 ay önce