正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

Our goal is to make the best chips physically possible for the large model needs of frontier labs. The MatX One chip delivers higher throughput than any announced product while also achieving the lowest latency. Reiner Pope joined Bloomberg TV this morning to discuss:

MatX

4,506 subscribers

40,049 次观看 • 5 个月前 •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 条评论

暂无评论

原始帖子的评论将显示在这里

相关视频

New interview: Reiner Pope, co-founder/CEO of MatX A counterintuitive throughput insight: “Low latency means small batch sizes. That is just Little’s law. Memory occupancy in HBM is proportional to batch size. So you can actually fit longer contexts than you could if the latency were larger. Low latency is not just a usability win, it improves throughput.” We get into: • The hybrid SRAM + HBM bet, and why pipeline parallelism finally works • Why sparse MoE drives MatX to “the most interconnect of any announced product” • Why frontier labs are willing to bet on an AI ASIC startup • Memory-bandwidth-efficient attention, numerics, and what MatX publishes (and what it does not) • Why 95% of model-side news is noise for chip design • The biggest challenges ahead 00:00 “We left Google one week before ChatGPT” 00:24 Intro: who is MatX 01:17 Origin story: leaving Google for LLM chips 02:21 GPT-3 and the “too expensive” problem 04:25 Why buy hardware that is not a GPU 05:52 Overcoming the CUDA moat 08:46 Early investors 09:35 The name MatX 09:59 The chip: matrix multiply + hybrid SRAM/HBM 12:11 Why pipeline parallelism finally works 14:22 Reading papers and Google going dark 15:20 Research agenda: attention and numerics 17:06 Five specs and meeting customers where they are 19:24 Why frontier labs are the natural first customer 20:32 Workloads: training, prefill, decode 22:18 Little’s law and the throughput case for low latency 24:29 Interconnect and MoE topology 26:35 Inside the team: 100 people, full stack 28:32 Agentic AI: 95% noise for hardware 30:35 KV cache sizing in an agentic world 32:11 How MatX uses AI for chip design (Verilog + BlueSpec) 34:23 Go to market: proving credibility under NDA 35:12 Porting effort for frontier labs 36:34 Biggest skepticism: manufacturing at gigawatt scale 37:32 Hiring plug Vikram Sekar

New interview: Reiner Pope, co-founder/CEO of MatX A counterintuitive throughput insight: “Low latency means small batch sizes. That is just Little’s law. Memory occupancy in HBM is proportional to batch size. So you can actually fit longer contexts than you could if the latency were larger. Low latency is not just a usability win, it improves throughput.” We get into: • The hybrid SRAM + HBM bet, and why pipeline parallelism finally works • Why sparse MoE drives MatX to “the most interconnect of any announced product” • Why frontier labs are willing to bet on an AI ASIC startup • Memory-bandwidth-efficient attention, numerics, and what MatX publishes (and what it does not) • Why 95% of model-side news is noise for chip design • The biggest challenges ahead 00:00 “We left Google one week before ChatGPT” 00:24 Intro: who is MatX 01:17 Origin story: leaving Google for LLM chips 02:21 GPT-3 and the “too expensive” problem 04:25 Why buy hardware that is not a GPU 05:52 Overcoming the CUDA moat 08:46 Early investors 09:35 The name MatX 09:59 The chip: matrix multiply + hybrid SRAM/HBM 12:11 Why pipeline parallelism finally works 14:22 Reading papers and Google going dark 15:20 Research agenda: attention and numerics 17:06 Five specs and meeting customers where they are 19:24 Why frontier labs are the natural first customer 20:32 Workloads: training, prefill, decode 22:18 Little’s law and the throughput case for low latency 24:29 Interconnect and MoE topology 26:35 Inside the team: 100 people, full stack 28:32 Agentic AI: 95% noise for hardware 30:35 KV cache sizing in an agentic world 32:11 How MatX uses AI for chip design (Verilog + BlueSpec) 34:23 Go to market: proving credibility under NDA 35:12 Porting effort for frontier labs 36:34 Biggest skepticism: manufacturing at gigawatt scale 37:32 Hiring plug Vikram Sekar

Semi Doped

19,439 次观看 • 3 个月前

$TSLA ELON MUSK: “One thing I’m trying to figure out is how to make enough chips, maybe we’ll work with Intel, but even whe we extrapolate the best case scenario for chip production form our suppliers. it is still not enough. I cant see any other way to get to the volume of chips we are looking for, so i think we will have to build a gigantic chip fab.” Tesla getting into the chip game 👀

$TSLA ELON MUSK: “One thing I’m trying to figure out is how to make enough chips, maybe we’ll work with Intel, but even whe we extrapolate the best case scenario for chip production form our suppliers. it is still not enough. I cant see any other way to get to the volume of chips we are looking for, so i think we will have to build a gigantic chip fab.” Tesla getting into the chip game 👀

amit

770,901 次观看 • 8 个月前

All of this past weekend's @WHCA dinner guests left unharmed – but the harm to our democracy is only worsening. I joined Joe Mathieu and Kailey Leinz on Bloomberg TV to discuss what we saw as attendees, and the implications it will have for the rest of America.

All of this past weekend's @WHCA dinner guests left unharmed – but the harm to our democracy is only worsening. I joined Joe Mathieu and Kailey Leinz on Bloomberg TV to discuss what we saw as attendees, and the implications it will have for the rest of America.

Frank Luntz

37,189 次观看 • 3 个月前

On a chip, even something as basic as selecting one input from a register file has to be built out of AND and OR gates wired together. Simple-seeming operations end up surprisingly expensive. Reiner Pope showed me how this stacks up. In older generations of chips, most of the die area was used for shuffling data around these circuits, not for the actual compute! This is one of the problems modern chip design tries to solve.

On a chip, even something as basic as selecting one input from a register file has to be built out of AND and OR gates wired together. Simple-seeming operations end up surprisingly expensive. Reiner Pope showed me how this stacks up. In older generations of chips, most of the die area was used for shuffling data around these circuits, not for the actual compute! This is one of the problems modern chip design tries to solve.

Dwarkesh Patel

27,876 次观看 • 2 个月前

🇺🇸 Trump: China wants American AI chips According to Trump, the topic of supplying Nvidia H200 chips to China came up during the discussions. “They have chips at a much higher level, but the H200 is a great chip. China needs it,” Trump said. He also noted that Beijing has not bought the H200 so far because it wanted to develop its own alternatives.

🇺🇸 Trump: China wants American AI chips According to Trump, the topic of supplying Nvidia H200 chips to China came up during the discussions. “They have chips at a much higher level, but the H200 is a great chip. China needs it,” Trump said. He also noted that Beijing has not bought the H200 so far because it wanted to develop its own alternatives.

NEXTA

38,160 次观看 • 2 个月前

Pope Leo challenges students participating in the Jubilee of the World of Education to work for a better society through education, which is “one of the most beautiful and powerful tools for changing the world.” “How wonderful it would be if one day your generation were remembered as the ‘generation plus,’ remembered for the extra drive you brought to the Church and the world,” Pope Leo said. But he also warned against keeping this desire a dream. The Pope explained the answer to achieving this is through education, “one of the most beautiful and powerful tools for changing the world.”

Pope Leo challenges students participating in the Jubilee of the World of Education to work for a better society through education, which is “one of the most beautiful and powerful tools for changing the world.” “How wonderful it would be if one day your generation were remembered as the ‘generation plus,’ remembered for the extra drive you brought to the Church and the world,” Pope Leo said. But he also warned against keeping this desire a dream. The Pope explained the answer to achieving this is through education, “one of the most beautiful and powerful tools for changing the world.”

Vatican News

20,116 次观看 • 8 个月前

Honored to be recognized in the State Capitol this morning. Many thanks to Sen. Massey and everyone else who made it possible. It is our goal to continue to make the great state of Tennessee proud.

Honored to be recognized in the State Capitol this morning. Many thanks to Sen. Massey and everyone else who made it possible. It is our goal to continue to make the great state of Tennessee proud.

Tennessee Baseball

23,034 次观看 • 1 年前

Over the past year, our #PerpetualPlanet Initiative partners have continued to push back the boundaries of what is possible, achieving remarkable breakthroughs. They are working harder than ever to make the world a better place. More #Reachforthecrown

Over the past year, our #PerpetualPlanet Initiative partners have continued to push back the boundaries of what is possible, achieving remarkable breakthroughs. They are working harder than ever to make the world a better place. More #Reachforthecrown

ROLEX

14,426 次观看 • 6 个月前

PM Netanyahu before departing for Washington: "I will discuss all the issues on the agenda with President Trump, first and foremost Iran. Our goal is to safeguard our security while expanding the circle of peace around us." He noted it will be his eighth meeting with Trump since the president's second term, more than any other world leader. He is also traveling to pay final respects to Sen. Lindsey Graham.

PM Netanyahu before departing for Washington: "I will discuss all the issues on the agenda with President Trump, first and foremost Iran. Our goal is to safeguard our security while expanding the circle of peace around us." He noted it will be his eighth meeting with Trump since the president's second term, more than any other world leader. He is also traveling to pay final respects to Sen. Lindsey Graham.

Open Source Intel

41,926 次观看 • 1 小时前

Jony Ive explains the Steve Jobs product philosophy that saved Apple from bankruptcy in 1998 “Our job isn’t to make money for Apple. Our job is to try and make the very best products we can.” Jony is aware that this sounds simplistic and easy to say given that Apple is one of the most valuable companies in the world today. But he points out that Steve Jobs stuck to this philosophy when Apple was on the verge of bankruptcy in 1998: “When Steve came back, that’s how he articulated what the goals of the company needed to be… It takes a tremendous courage when you’re losing fabulously large amounts of money to say ‘Our goal isn’t turnaround. Our goal is to make a great product.’ That’s not a natural reflex to that situation. The reflex is: let’s not spend this money and let’s try to get a little bit more because we’re about to go out of business.”

Jony Ive explains the Steve Jobs product philosophy that saved Apple from bankruptcy in 1998 “Our job isn’t to make money for Apple. Our job is to try and make the very best products we can.” Jony is aware that this sounds simplistic and easy to say given that Apple is one of the most valuable companies in the world today. But he points out that Steve Jobs stuck to this philosophy when Apple was on the verge of bankruptcy in 1998: “When Steve came back, that’s how he articulated what the goals of the company needed to be… It takes a tremendous courage when you’re losing fabulously large amounts of money to say ‘Our goal isn’t turnaround. Our goal is to make a great product.’ That’s not a natural reflex to that situation. The reflex is: let’s not spend this money and let’s try to get a little bit more because we’re about to go out of business.”

Startup Archive

199,892 次观看 • 7 个月前

Do you think this is the best video of yasmina? Thanks to our friends to make it possible. 🙏🏽

Do you think this is the best video of yasmina? Thanks to our friends to make it possible. 🙏🏽

Yasminabradyclub💗

43,581 次观看 • 7 个月前

Do you think this is the best video of yasmina? Thanks to our friends to make it possible. 🙏🏽

Do you think this is the best video of yasmina? Thanks to our friends to make it possible. 🙏🏽

Yasminabradyclub💗

44,781 次观看 • 7 个月前

The top hacker in the US is not a human, but a machine. XBOW founder and CEO Oege de Moor and Apoorv Agrawal of Brad Gerstner joined Bloomberg TV this morning to talk about the milestone.

The top hacker in the US is not a human, but a machine. XBOW founder and CEO Oege de Moor and Apoorv Agrawal of Brad Gerstner joined Bloomberg TV this morning to talk about the milestone.

XBOW

58,665 次观看 • 1 年前

There's an economics theorem called Alchian-Allen. And it has the very interesting implication that AI labs will be able to charge *higher* margins on their best models as compute gets scarcer. As compute gets more expensive, the cost of running any model goes up. So you might as well pay a bit more to make sure you're running the very best model. Which means the economics of being at the frontier improve, because if you’re not running the very best model, then you’re underutilizing this very precious compute. This pushes the AI model market towards winner-take-all; if you're the best, you can get away with charging an even higher margin. Dylan Patel tells me that we’re already seeing this today: all the revenue in the industry is on the best models. That’s the Alchian-Allen effect. If there’s a cost increase that’s roughly the same for all products, then the relative difference in price between higher and lower quality goods actually goes down. Consumers become relatively more willing to pay for the premium product. And it means that as the compute shortage hits, AI labs can capture more margin - not less, as you might expect - because consumers are choosing premium models more often.

There's an economics theorem called Alchian-Allen. And it has the very interesting implication that AI labs will be able to charge higher margins on their best models as compute gets scarcer. As compute gets more expensive, the cost of running any model goes up. So you might as well pay a bit more to make sure you're running the very best model. Which means the economics of being at the frontier improve, because if you’re not running the very best model, then you’re underutilizing this very precious compute. This pushes the AI model market towards winner-take-all; if you're the best, you can get away with charging an even higher margin. Dylan Patel tells me that we’re already seeing this today: all the revenue in the industry is on the best models. That’s the Alchian-Allen effect. If there’s a cost increase that’s roughly the same for all products, then the relative difference in price between higher and lower quality goods actually goes down. Consumers become relatively more willing to pay for the premium product. And it means that as the compute shortage hits, AI labs can capture more margin - not less, as you might expect - because consumers are choosing premium models more often.

Dwarkesh Patel

322,651 次观看 • 4 个月前

Before making any structural changes across all sectors, the economy needs to be stabilised. For this, the govt intends to continue with the IMF program & proceed with the debt restructuring, with the aim of achieving debt sustainability as soon as possible -President #AKD

Before making any structural changes across all sectors, the economy needs to be stabilised. For this, the govt intends to continue with the IMF program & proceed with the debt restructuring, with the aim of achieving debt sustainability as soon as possible -President #AKD

Market News🇱🇰

121,659 次观看 • 1 年前

Our Founder and CEO JoeBen Bevirt: "Our goal at Joby has always been to make traveling by air as accessible, clean and quiet as possible. Electric propulsion will be key to achieving our mission, but the energy density of hydrogen is 100 times greater than today’s batteries, and it has three times the energy density of jet fuel."

Our Founder and CEO JoeBen Bevirt: "Our goal at Joby has always been to make traveling by air as accessible, clean and quiet as possible. Electric propulsion will be key to achieving our mission, but the energy density of hydrogen is 100 times greater than today’s batteries, and it has three times the energy density of jet fuel."

Joby Aviation

17,659 次观看 • 1 年前

"It was the best first day of trading for any of our ETFs.. We had to start with bitcoin (ETF), but this is just the first of a long road map of new products both on the asset mgt side and wealth biz" - Any Oldenburg head of Digital Assets at Morgan Stanley on Bloomberg TV today re $MSBT

"It was the best first day of trading for any of our ETFs.. We had to start with bitcoin (ETF), but this is just the first of a long road map of new products both on the asset mgt side and wealth biz" - Any Oldenburg head of Digital Assets at Morgan Stanley on Bloomberg TV today re $MSBT

Eric Balchunas

51,928 次观看 • 3 个月前

Not every AI interaction is best served with a large frontier model. There is a long tail of trillion-token use cases, from tagging to search, best served by a sub-10b-parameter model that runs in milliseconds and costs many orders of magnitude less than the frontier. We built Freesolo Flash to make training loops like SFT and RL easy and end-to-end completable through your coding agent.

Not every AI interaction is best served with a large frontier model. There is a long tail of trillion-token use cases, from tagging to search, best served by a sub-10b-parameter model that runs in milliseconds and costs many orders of magnitude less than the frontier. We built Freesolo Flash to make training loops like SFT and RL easy and end-to-end completable through your coding agent.

Freesolo

26,686 次观看 • 9 天前