Intel's Crazy Plan for AI Chips IS WORKING! (Supercut)

Hi Sam, thank you very much for joining us, of course. Hey, a couple quick questions. You know the first one is 7 billion. What is this all about? Look, I'm sure everyone knows that they shouldn't believe everything they read. We in the press really believe that the core of the truth is that we believe that investing a lot of money in AI computing power data centers will be important in providing the number of services that people want and the tools that we all have. I really think we're going to get a lot of value to help create better futures, mhm, so there's a lot of investments to be made here, I think so, but it's about a lot of things, isn't it, I mean, it's like a world. investment to help many, many different efforts, many different people and not just

chips

of course but all the infrastructure around it, okay, second question, you and I are a little bullish on technology, so why Should we be optimistic in technology regarding AI?

I think this is going to be one of the best tools that humans have invented so far if you think about what we can do with more

intel

ligence available to us U more abundant

intel

ligence cheaper prices things that humans alone are not capable of doing uh we will all be able to invent new things for each other that amaze us that's how the future gets better by making things smarter I like when you say that, uh, third question, our founder Andy Grove said the famous phrase: the paranoid survive, very paranoid about yourself, your success, but also paranoid about competition and things that could go wrong and there are those who are a little paranoid about AI today, what do you tell them?

I think people are right to be cautious about AI. I think this will be an extraordinarily powerful technology. I think it's we're going to reshape important parts of the way we live and work and the way our economy works and part of our strategy, what we call iterative deployment, is that we want to get these things out into the world early so that people can adapt, think and provide information. and that we can be ready for the change man that will come with time, so I think it's great, yeah, getting them out early, people try them, that's how things get better, yeah, making them completely better, well , Sam, you have honored us by joining.

Intel Foundry is dated right, the partnership we have with you is tremendous, we're going to do great science together, great AI machines together, and fundamentally make the world a better place. Thank you very much, thank you for keeping me alive and well. until the periodic table runs out we are not done with Mo's law we bend physics we create molecules we do amazing things and Moore's Law is alive and well but we have to do it a different way we have to enable reliable and sustainable supply chains globally resilient and at the end of our conference today I'm joined by Sam Altman and he's going to challenge the capacity needs of the industry and the chip system needs new capabilities, new testing and packaging capabilities and that's what a foundry does. systems and how we describe them.

In this vision of becoming the number two foundry in the world, we realized that there are only a few companies that can do this and that have the capital capacity, the research and development and the longevity to do it and now that we are in an economic downturn, welcome to semiconductors and then the explosion of AI and the cyclical nature of the industry and we have seen geographical instability and active wars in Israel and Ukraine and tensions and Taiwan in the Strait, this is anything but a chain of resilient supply today and that's why it becomes a large, significantly, the second largest foundry, but become the most sustainable foundry in the world, the most resilient foundry in the world, because that's exactly what the world needs, but surprisingly As you know, in 1990 80% of semiconductors were manufactured in us and in Europe, today 80% in a small concentrated area.

In Asia we have seen this long and constant decline in terms of our supply chains to the world, nothing should depend on a single port to a single country, a single place in the world, we need resilience, resilient access to supply chains and capacity in the right regions at the right time and therefore the choice the opportunity to drive systemic change in where and how we drive the most important aspect of our future where the technology is Supply chains what aspect of your life it's not becoming more digital well everything is your healthcare your financial you are social and with that we just call it silicon and the economy merge in an inextricable way and as administrators of Moore's Law we see this relentless search for more efficient Computing , more capable, more scalable and so we have been on this journey, we also announced that we are going to get five nodes in four years, we are going to do something unprecedented in the industry to return Intel to the leadership in process technology and while we are not finished today, we see that the end is soon in front of us on that journey and the shipment, increase and volume of Intel 7, Intel 4 with our launch cor ultr, shipment, increase and volume of Intel 3 is certified production and we will be with our server products.

It will be released in the first half of the year and it will be produced in volume, so we have been on an incredible journey, but then it continues into what we call the angstrom era and for this Intel 20a and Intel 18a, the adoption of tape fueled a new structure of transistors. of energy through power supply technology, the embrace of the first major new transistor in architecture since the year 12, this is a Mona Lisa, no, no, this is a rembrand, no, no, I think it's a Michelangelo, sculpted in silicon, because these truly are works of art and I'm excited about the progress and with the 18a finishes, we've already sent Fab our first 18a products and this is a test tip for Clearwater Forest, so I'm excited , this is what we call a family photo for the children to come with dad. here we go five nodes in four years and I want to announce, describe and give a moment to our latest 18a customer, my friend of decades, Sacha Nadella, speaking on behalf of Microsoft as the newest 18a customer, let's hear from Sacha now, thank you very much , Pat.

It is a pleasure to accompany you at his launch event. It is clear that we are in the midst of a very exciting platform change that will fundamentally transform the productivity of each individual organization and the entire industry. To achieve this Vision we will need a reliable supply of the most advanced, high-performance, high-quality semiconductor products and all of us at Microsoft are committed to supporting Intel's efforts to build a strong supply chain here in the United States. That's why we're so excited to work with Intel Foundry services and why we've chosen a chip. design that we

plan

to produce on Intel's 18a process.

We look forward to sharing more details in the future and I can't wait to see everything we will offer together to our customers in the years to come. Thank you so much. Sacha, can I tell you more about what's coming after 18a? Today we announced that we are expanding these nodes, we are adding uh major and minor nodes, a combination of older and cutting-edge nodes to ensure that our customers have access to the process technology they need today we are announcing Intel 14a for the first time, think about this as four 1.4 nanometer technology, but Intel 14a ventures deep into the angstrom 14a era first processing, but we are also announcing that we are expanding our nodes as You see here adding P noes improvements to the existing ones adding performance adding t-noes uh across silicon through new feature enhancements with e noes on the roadmap completing that capacity roadmap today we first announced Intel 16e There are also improvements to our Intel 16 technology, so that we are completing that full set of nodes, the roadmap that we have to go further, but as we have seen throughout this period of time, this explosion of the AI ​​era.

These cool packages have gotten really cool, which is why Intel Foundry now offers an extensive suite of advanced assembly and testing. Intel Technologies 25 years ago drove the standardization of organic packages, now we are driving the next generation of glass-based packages and with that the ability to interface directly with optics and waveguides directly in the package construction for system capabilities more advanced, as you know, and Intel Foundry has added a number of additional AI clients to our portfolio of package offerings, plus the AI ​​era needs advanced wafers, but it even needs more systems and packaging capabilities, an area in which Intel is the undisputed leader and this now includes as customers some of the world's largest AI leaders, as I include, my time on stage today is a day 3 years in the making and I couldn't be more proud of the Intel team that is supporting this rebuild of this iconic company and today you will hear several of those leaders bring together the foundry capabilities of the world's first system for the AI ​​era and with that it is my pleasure to introduce a friend to the stage for decades and to the zealous leader of our Foundry Services, none other than Stu pan.

Please welcome Stu to St. Our people can feel and sense this urge, can hear it in their voices, can see it. In their actions they know that it is not necessary to be in Taiwan to build the most advanced semiconductors in the world. They have made these five NOS in four years. This audacious goal that Pat set. You can see the checkmarks in all the boxes here. We are ready to earn the right to be your Foundry provider, no one is going to give us that, we want to earn it, so that's one perspective. I want to offer a second perspective and this comes from Chris Miller, the author of the book Chip War.

This is a New York Times bestseller, Financial Times book of the year, it took Chris 10 years to write it now Chris in a speech in October said this quote to employees said quote Intel is the most important company of the last 50 years so I called Chris and said are you okay if I say this publicly? Yeah, okay, go ahead, good publicity for the book. I came back to Intel 3 months after Pat did it because I believe this quote is true and I believed it long before Chris said it and what you're doing. I'm going to hear over the next 40 minutes or so all the reasons why Chip War season 2 is about to begin, so let's start from the beginning.

Gordon's original article was called "cramming more components on Integrated circuits" and was published in 1965 every time I read it. I'm surprised by how insistent you were in the last section of the article. Gordon comments that we reached Doomsday, quote, it can be cheaper to build large systems from smaller functions that are packaged and interconnected separately and this is why I think we are in the area of ​​foundry systems . You can no longer make only monolithic devices. You have to divide them. The thermal profiles must be adapted. Costs must be addressed. You have to have flexibility. And all this is what drives it. a new level of systems thinking In The Foundry business, in this era of systems, it is not only necessary to have open standards between devices, but it is also necessary to have standards on the device.

Consider the fact that running a daily training model requires 100,000 CPUs running and in concert. all on the same data set, the next round of training models will require a million CPUs, that's something that requires standards everywhere, from the internal

chips

to the external chips, this is what it takes for our customers succeed and this is the fundamental reason why Intel is a systems company that is transforming. in Foundry and not the other way around, so now that we've described systems thinking, let's look at how the original Foundry model was created now that CC Way has been kind enough to mention us in their latest earnings calls.

I thought I would return the favor and talk about tsmc in my presentation. This was a presentation that the Morris chain gave at MIT last October and in it it describes the Foundry model, it says that it is research and development, it is water manufacturing, it is advanced packaging and it says in the red line that that is what tsmc does and says on the blue line Intel everyone else does everything else you could even call an IDM 1.0 type of manufacturer now tsmc has had incredible success with this model discipline execution discipline consistent strategy Innovation but, to quote Bob Dylan The Times , they're changing.

I tell you why there is an idea of ​​co-optimization of systems technology in which the workload of applications is analyzed Software System architecture Memory interconnects all these different things. What happens today is that people focus on their layer and maybe on the layersuperior, in fact this. It's really what a classic Foundry does today, but now we're in the realm of the exponential. Sam alman will come out later this afternoon and talk about how he doesn't have enough ability to do what he wants to do, you have to look. In all of these combinations you can do a couple of things right, but making a system that coordinates the activities of solving a training model on 100,000 CPUs requires that you get all the gear ratios right if they don't match in memory if not If they don't match In networks, you end up wasting valuable cycles and resources, so we have to get 100 times more out of what we're doing, so we describe the evolution.

Now let's develop the strategy and talk about what is the revolution in all this that we like. Think of this in three basic layers. First, we have to be a world-class founder. Rowan Chen, Qualcomm's COO, talks about the fact that silicon talks and silicon talks in four different ways: performance, power, area and cost, without that it's not. In business, the next layer of the pyramid is all about resilient and sustainable supplies. This need to have capacity around the world to build this and do it sustainably. The top layer of the pyramid is this idea of ​​chip systems, we're putting a system on a chip and we're

working

with our foundry partners to create chip systems, so let's explain to you what that means.

I talked earlier about this Day of Judgment, what that means is happening now that we can't anymore. We do designs at the monolithic level, now we are in limited retical designs, design sizes that are 800 square millimeters of silicon because we have to go beyond the Ral limits and the thermal constraints and because of the uniform cost constraints because when these sizes are built of really large arrays are really expensive, isn't there a better way to take advantage of them?, as Pat talked about taking smaller tiles in the more advanced notes, getting better returns, packaging them together, having more flexibility to do everything you really require if you want a system. number of chips so you can see in the animation here how we built it and this is literally how we are going to build in the factory the idea of ​​a substrate, the idea of ​​Base, the idea of ​​logic tiles, the idea of ​​IO tiles on the side why doing this gives our clients the ability to optimally exchange what they need for their design.

The things you do for a training engine will be different than what you do for an inference engine and just by having all these levers. To go shooting, can you do this? We learned a lot through a device called ponovo or what the people at The Branding call the Intel GPU Max series data center, it's ANC, it's 100 billion transistors, so it's dozens of tiles of chiplets, 47 of them are multiple vendors the way we coexist with tsmc in the same package, right, we develop testing techniques to do that, we do this singular die testing idea, what does that mean?

It means that every die that goes into that packet is a known good die, why? important because you want them all to be good, the assembly test shield on this device is 95% and it's also the Super Bowl of integrated design. Now, what can you do right with it if you're Argon National Labs and our partners at HP? You build a really big supercomputer and they built a computer that had 66,000 Pacos 20,000 Sapphire Rapids and it looks like this, what do you do with something like that? Well, you solve some really difficult scientific problems if you want to model airflow through the wing, you can do that on a workstation if you want to model airflow through an air

plan

e, you do it on this if you want to model air reactions.

Fusion, which is quite difficult stuff to model, you do it in a device like this if you want to model if you want to model drugs that cure cancer at a molecular level, do it like this it's 600 tons of computation it's four tennis courts it's the weight of an Airbus It has 300 miles of optical cable it takes 34,000 gallons per minute to cool the way the faucet in your house is one gallon per minute, 34,000 gallons per minute, so when we talk about how to design things, we have to find ways to do it with greater energy efficiency so that it is more profitable.

Today this is the second fastest supercomputer. By the way, there are approximately 100,000 CPUs in the world to handle the demands of AI, we are going to go much further than that and that is why this idea of ​​scalability is so important. Now let's go to the top of the pyramid, this idea of ​​chip systems. Chip systems require, as Pat mentioned, excellent packaging technology. We have created a standard called UCI that enables chip-to-chip connectivity. Think of it as the PCI Express of what we did in the mid 90's. This idea can be added, mixed and matched by the way. you can mix and match different Foundry providers, we like that because we're kind of underdogs in all of this, some of our competitors not so much, but our customers want this kind of flexibility, as you can see in the render what we're doing. is to take systems on a chip and build chip systems, why do you need chip systems?

Because it's the same problem I mentioned above. You need to train models with 100,000 CPUs now and potentially a million in the future and maybe up to 10 million when If we're moving coordinated data across all of these devices, you need to have standards and connectivity and that's what we provide and, really, when You think about it, what will it take to bring AI everywhere? How can you make it profitable? you make it capital efficient, yes, yes, I know Sam is asking for billions, but we want to make sure he's spending all that money as profitably as possible, so let's look at that: you have data center chips that double year after year, but the efficiency needs are the things that are really striking The New York Times published an article that AI could soon need as much electricity as an entire country, so I'm sure you're curious which countries Sweden, the Netherlands, Argentina, if you were to run all the AI ​​servers that market estimates have on djx2 dx100, those types of devices would take 85 to 134 terawatt hours (terawatt hours) by the way, the great state of California, its Total power generation capacity per day is 30 hours per hour for the entire state, so bringing AI everywhere.

This will require our Foundry partners to figure out how to do this profitably, this is what this roadmap will require and I think this is what sets us apart from other Foundry approaches that you should start with first. Table Sticks, as I mentioned, has to start with a great process added to that package, which we think is a unique difference for us because we build a lot of server parts and we're taking everything we learn from our server business and offering It Depends from our Foundry partners, this is up for grabs and this is what it takes to be a Foundry player.

We have this planned literally in all these dimensions for the next 5 years. Let's talk about cooling. I mentioned that argon needs 34,000 gallons per minute to cool it the next wave of devices will have to be submerged and cooled and today the Intel Zeon line of products is the only product that offers an immersion cooling guarantee immersion cooling allows us to deliver power much more effectively in a data center that we are in. We'll take what we've learned there and offer it to our Foundry customers so that when we start looking at 2000 watt devices in 5 years we have a way to cool that memory. maybe we'll put memory in the device to make it more computationally cost effective and over the next five years you'll hear us talk a lot about new technologies and new ways to increase memory bandwidth, why decrease the need energy consumption a five? -year type of appearance interconnects interconnections between chips the idea of ​​​​having high speed series interconnect all the things that you expect from a Foundry supplier is what we will give you and we are not going to plan just what is there today, have Keep in mind that as a standards company we do this on all the different standards and it's our job to make those standards available to all of you and lastly, think about net

working

, think about a network card that's capable of handling the AI demands that it models what you have to do with Ethernet today is not good enough, so we work closely with various Ethernet standards partners to discover ways to make Ethernet more capable of developing the idea of ​​systems. of chips and ethernet provides that low cost high bandwidth potential to exploit and do that add to that phonics add to that Technologies beyond that here again is another 5 year roadmap this morning we are announcing a new partnership with The Arm Fusion Business Initiative How do we take advantage of all the programs Arm has to offer? bringing design capability design education to all your clients we are doing this with arm we will do co-investments we will do joint programs we will provide shuttles at scale arm we will provide intellectual property at scale and that's where we are I'm going to drive this next wave of innovation and it's really exciting, so now I would like to invite R to come on stage with me.

Hello sir, it's a pleasure to have you here, so Renee, first of all, what kind of universe would you ever be in? I thought you'd see Arm and Intel together, but we quickly realized that 80% of the way for tsmc runs has an armed device. There is no way you can be in The Foundry Business without a partnership with Arm. and I think you'd probably like to tell us about some of them, starting perhaps with the announcements you made this morning. Yeah, well, thanks for having me, like you said, these are kind of strange bedfellows. I was trying to think of a parallel. "I could tell the relative story and the only thing I can think of is when Walt Mossberg asked Steve Jobs what it was like to see iTunes running on Windows and I think he said it was like an ice drinker in hell, but I didn't." I'm not going to go that far.

I won't go that far. The announcements we made today were about our Neoverse product line, which is the product line we use for the data center that just blew up. It was on a very high growth trajectory. Before the AI ​​wave and now it's gotten even stronger, the Nee V3 that we announced today, which is 50% faster than the Neoverse N2, and then our N3, which is 20% faster but also much faster. efficient, and if you think about these AI data centers. that are generating hundreds of megawatts and more efficiency matters, so yes, today was a great day for the arm that we are working on cutting edge technology when talking about 18a and the system that sinks into the packaging, this is the spearhead in terms of innovation and you have been fantastic, your groups have been a real pleasure to work with, the level of engineering commitment, the depth of the technical discussions, the information we get, we couldn't announce the partnership we have .

We've forged without it, we talked a little bit about Faraday, let's talk a little bit about what the Faray world plays in Neoverse and CSS and why what maybe find out what CSS is for and why it's so important to everyone here in the audience, yeah , so we announced a new strategy a few months ago around what we call computing subsystems and the way to think about this is essentially, instead of delivering blocks of IP, a mesh network, a memory controller, the CPUs themselves, we deliver a complete subsystem. verified completed, meaning if you want 64 cores, 96 cores, 128 cores for a CPU, we deliver everything in terms of that system validated, verified and it will work now.

One of the great benefits of this is simply the fact that time waits for no one, classic. classic quote, the amount of time it takes to design these s so's is really very difficult and it's really very long if we deliver the final IP to a client they still need to put all those pieces together if we can put all that together for them before on the same date they would have obtained the IP block, that's a huge benefit and then when you add to that, the processing cycle times are getting longer and longer. I know you guys are doing your best, but you know better euv.

Steps mean a more complex time through the factory and that simply means processing times are ultimately long. You have complex packaging, you know, that adds a lot of time, so anything you can do up front to benefit design time is really beneficial. we announced this program called uh arm total design that Intel is a partner of and in that model customers can go to their partners and work with those people to get their design out and that's what Faraday is, so Faraday is basically doing that. able to match your IP withour blocks and something that just ends the quote works and everyone wins, the product comes out faster, it's compliant and will work great, or René, thanks for coming and joining us today, a pleasure, I totally appreciate the partnership and everything. we are doing together everything we are going to do together thank you again thank you all so to finish my talk I would like to go back to where I started this is an ad filmed in the 1997 Super Bowl the year the Green Bay Packers beat the Patriots , the halftime show was by The Blues Brothers and James Brown and was eight years after Taylor Swift was born.

I had to do it. You can't associate supers without Taylor Swift and this commercial was iconic, but you know things have changed a lot. in recent years and that change will only accelerate as AI impacts your work your business your life how AI impacted the design cycle AI improves your engineers' ability to do much more with much less effort, so we decided The point with this commercial was to take a vision of generative AI, so this commercial didn't buy people who were harmed. We did it 100% with AI. It was made by very creative director Dave Clark from Los Angeles and he is a big proponent of using AI tools to not take away creativity but amplify creativity, so as you leave our session today remember this: the only imitation of what What we can do with AI is your imagination, thank you.