Fortune 500

Get insights into GPU including business profile, stock information, news, and rankings from Fortune.

GPU (GPU): Company Profile, Stock Price, News, Rankings

legacy

Companies

GPU | Fortune

GPU (GPU): Company Profile, Stock Price, News, Rankings | Fortune

This week Jeff and Leo discuss ex-NYC Mayor Eric Adams’ foray in the crypto space, possible insider trading on Polymarket, and the case for concern when it comes to World Liberty Financial’s banking charter application. Leo also talks with Lighter CEO Vlad Novakovski about decentralized exchanges and the bull case for crypto.

 A rugpull from NYC’s ex-mayor, prediction market hijinks & Lighter’s CEO | Fortune’s Crypto Playbook

Fortune_Crypto

Wayve CEO Alex Kendall takes Fortune’s Jeremy Kahn for a ride in London, showcasing the mechanics of the company’s AI-powered vehicle running on a single GPU

Microsoft, NVIDIA, and SoftBank have bet over $1 billion on this self-driving startup—here’s what it’s like to take a ride

Features

Microsoft, NVIDIA, and SoftBank have bet over $1 billion on this self-driving startup

Just engaged autonomy and look at that. The A I is now controlling the speed, the steering, everything about the vehicle and we're straight into traffic. Let's watch how we safely navigate around that cyclist. You can see we're slowing down, giving us some distance. So we're going to make a left turn, which would be interesting to see. So, yeah, we should slow down and see how we let the cyclist go through and then make the left turn when we moved over to make a left turn and we had to pick which lane we were going to, was it wanting to that? It knows from mapping that, that it has to be the left lane or did it actually read the indications on the roadway? So it gets to simply navigation from like drive straight or take a lift here or things like this a bit like what you might get if you use a set net system, some of the way the car is control the speed it chooses or the line it takes. None of this is hand programmed, not following a map, but it's making all of these decisions based on what it sees and the whole driving models on a single GP U in the car. Yeah. Yeah, there's nothing on the cloud. It's all on board the vehicle on a single GP U. A similar amount of computes to what we're seeing in production vehicles today. How did you gather the training data for the system? We use a wide variety of sources of training data to make sure that we learn the most safe and performance system. So of course, we have our own fleet that we test throughout the day and night, getting experience in London and we do some trips to other cities as well. From time to time, we partnered with ASDA and 02 of the largest grocery fleets here. Together we collect data from their grocery fleet and we have some other initiatives like that, that we, we're sort of taking off around the world. You mentioned, it's sort of like a large language model takes in all of that data, including things that you wouldn't necessarily want the model models behavior on. But having that amount of data, it sort of makes the system more robust in the end. But then of course, with large language models that you do have that sort of reinforcement learning through human feedback process. Is there some similar equivalent thing that you do with the data? Yeah, it is. And the fact that the way that we test and we give the feedback or in simulation when we resim and look to understand how the system performs. It's, it's a, it's very similar form of feedback in general. The way that, you know, I've been, I've been working on these A I models for over a decade now and I've seen breakthroughs in computer vision or robotics or natural language. They all tend to converge we're seeing now the same techniques apply in different fields. So the rise of synthetic data, for example, in large language models, we see the same thing with generative A I and what we can do in simulation for driving multi. Now we have vision uh language models, you know, say with the LLN similarly, we have lingo which gives us language capabilities in in this car. So all of these kind of trends are playing out in a very similar way. Of course, the interesting thing in this space is that we have the integration problem of making it run real time in a car and the safety challenges of running in the open world where we can't control what might happen. Ultimately, I think what we're learning is that A I is breaking through a lot of the glass ceiling that we've seen in the industry so far, removing the need for maps and being able to leave the geo fence, enabling the use of cost-effective hardware and sensors. Uh But more importantly, solving the long tail of cases and producing behavior that's intelligent enough to drive in complex environments like London

Nala Robotics is using A.I. to automate jobs in restaurant kitchens.

Robots are moving into restaurant kitchens thanks to A.I.

Tech

a robot is nothing but a mechanical reflection of us. As long as it's given the right mode of payment, the data is being handled properly. Robot actually can prove equal, if not better than a human, other robotics uh is uh ai based technology company. We are specializing in kitchen robotics, uh mainly automating and introducing Ai concepts into the back of the kitchen. Most of the technological advances that happened in the restaurant industry are the front of the house uh has been introduced for awhile, mainly in the system. We realized that the back of the house does not have any many technological advances. And that's when we started introducing robotics. Nala can actually perform two important tasks. One is, it can perform the task of line cook as well as the chef. The robotic arms actually get the recipe and actually perform the tasks that are normally performed. Then the chef part of it is the AI element. The software is the element we have our own Os called uh that's the brain of the, that's more of a chef that can create new recipes, come with new ideas, you know, look at the ingredients and then say, hey, you know what, why don't you make a new dish today? So that's a part of it. It is multi cuisine. So it actually can virtually cook anything that is programmed to do given a few ingredients. It actually can come up with new space on its own uh it actually would use feedback mechanisms that the customers give to enhance the taste and come with with new new cuisines. In fact, The National Restaurant Association has actually surveyed restaurant operators across the country not only about their their thoughts about technology, but what they're using and what they expect to do in the future. And as you can imagine, they view technology as a very, very important competitive differentiator. I was looking into robotics and I always thought it was gonna be like 3-5 years down the road and then meeting AJ and his company. He changed my mind real quick. And uh yeah, it's here, it's it's now and it's happening. Pizza Yola is physically designed for classic american food. So it's not just making a pizza. It actually can make fried wings, fried chicken, it can make salads. Pizza Yola is 12 ft by 12 ft. It's much smaller than the system that can fit into any any pizza chain right now, what's unique about us is everything we do is larger than standard. So everything he has to build for us was custom. We are going to retrofit the first ones to fit some stores eventually will perfect the model which then the kitchen will be built around the pizza yola. When you apply technology to an extremely, extremely labor intensive industry, which the restaurant industry is the typical operator can get some substantial productivity and efficiency gains as we scale into certain markets, it's hard to find skilled labor with Saleh's product. The training will be much more efficient right now labor costs have significantly gone up, California is introducing a new law, increasing the minimum wage even higher. And that that's hindering the operations of a lot of restaurants in fact, so uh can address that ultimately would do, it will replace the jobs in our kitchen that most of our staff don't want to do and also be able to work the hours that most people don't want to work for. Late nights and the weekends. Our customers reaction is one of the biggest concerns. We want to be very clear that this product is not taking any jobs away from anybody, it's there to assist and to help make their lives better and to operate more efficiently. The cost of computing has come down significantly. The improvements in Gpu. The improvements in in the computing world has significantly improved the quality of current Ai machines. In fact added to that the cost of robots have significantly come down, robotic arms are more affordable than what it was 10 years ago that the advances that happened in the fields of vision in fields of you know, speech recognition made it all 10 years from now. I don't see restaurants not having some type of Ai assisted robotic in their in their operation. The industry has always demonstrated its innovativeness and ability to rapidly respond to consumer changing wants and needs. And so the ability to to continue to innovate and rapidly adapt. Including the integration of technology um becomes pivotal for their success and growth going forward.

Dario Gil, Director of Research, IBM 
Interviewer: Brian O’Keefe, FORTUNE


Brainstorm Tech 2022: When Does Quantum Become Real, And How?

Conferences

When Does Quantum Become Real, And How?


Brainstorm Tech 2022: When Does Quantum Become Real, And How?


All right. I want to start with a quick show of hands how many people in this room think you can explain quantum computing. Nobody's raising their hands. How many people here have at least heard of a cubit? Alright, alright. That's pretty good progress. Yeah. Not the video game, the computing thing. Um So uh sorry, this is a pretty savvy group but they're still brave enough to admit they don't really understand quantum computing. So let's turn to you the expert first. Let's start with uh Maybe you could just give us a quick refresher course on what quantum computing is, what it means, how it works. You know how it's different from classical computing. Just like give us like that quick overview so we can dive into sort of the details. Well, the first thing to understand is that quantum computing is the first time that the category of computing and I mean computing with a capital C. Has branched. So that's not an everyday occurrence that that happens. And we moving forward, you already refer to the term that we will adjudicate to all the computers that we built to date, which will be called classical computers. And then there's this new branch called quantum computers. And it's you know, nothing less than a re foundation of the world of information technology. If you look at how we have built the modern information technology edifice, it actually began back to uh you know, fundamental idea that was put forward by live in. It's the great philosopher and mathematician co inventor of calculus with Newton. That envision envision it was, you know, creature of the renaissance. This idea of creating perfect logic. And he understood that you could encode almost all information in a in binary digit And the idea was you could reason over it and you could create this perfect reasoning machines. He didn't know how to build them. So it took until the 1940s with Claude Shannon and then on touring and others to start building and creating those ideas. And we introduced the term the binary digit, the bit, the zero and one. And in that abstraction we understood that the zeros and ones could be separated from their physical manifestation meaning hey, you could build a computer with vacuum tubes or you could build it with semiconductors and with transistors. And we've made, you know, the story of moore's law has been the story of making those bids more more efficient and cheaper and so on. At this separation between the zeros and ones and their physical manifestation bothered a number of physicists. And they started probing into this question. And it turns out long story short is that there is a relationship between physics and information and there's a set of properties from quantum physics and from physics in general uh called superposition entanglement and interference that get mixed with this idea of information such that we can represent it much more richly meaning instead of just having two states and build machines that process those two states. Now the information is more rich. The representation is more diverse and quantum computers are nothing less than the ability to create these new expanded representation of information to create these quantum bits that have an exponential number of states present in them. If you build the machines properly and to manipulate that information in such a way that you can perform computations that you simply couldn't do classically or to do them much more efficiently. So new generation of machines that have re founded the very idea of information. It doesn't even preserve the idea of the zero Under 1. Okay, so let's get quickly now into the central question of our discussion, which is when does it become real? And let's start with just is it real now? I mean like you know, I I've seen you know stories about mixed results in the comparison of quantum computing results versus classical computing in terms of where we aren't now with quantum computing but we have obviously IBM but google and Microsoft and amazon and a lot of inter backed startups investing a lot of money in this field. There seems to be a lot of their there. But what can we look at right now to see that uh that quantum is actually starting to go from theory to you know, brilliant theory to real world manifestation and and what are sort of some of the signposts that we're seeing that mark that progress. Well, first of all quantum computers are real and they exist right now. Just to give you some some grounding at IBM we have over 20 computers to 20 quantum computers right now on IBM cloud at clients and uh and the community can use every day all over the world again to give you a measure. 3.5 billion quantum circuits are run every day on IBM quantum computers. Since we put the first quantum computer on the cloud, it was May 2016 over one trillion quantum circuits have been run on actual quantum computers. We have gone from the first generation that we put out had just five cubits. We went from 5 to 27 last year we broke the 100 cubit barrier. We have now a system that has 127. This year we will deliver a system with 433 cubits. Next year there'll be one in 2023 with over 1000 and by 2025 or plan is to build a machine with over 4000 cubits. So first of all the reel machines, you can program them, you can run things on it. Over 1000 scientific publications have been run on IBM quantum computers as an example. So it's a real technology that is present. I would say the experimental aspect of building real devices and real quantum computers has been the food of the last 20 years But really has been the last decade and specifically the last 5-6 years will be seeing this very, very sophisticated advancement now at the same time no one has yet built quantum computers that for a practical application can outperform what you can do with a classical computer. That threshold, what we call quantum advantage is something that you know we're all hard at work on the field and that is expected. That certainly is going to happen this decade. But it's likely that some first demonstrations can happen as early as the next few years. So that's the big next milestone in the in the field. But there's now hundreds of thousands of people who are writing programs on quantum computers. Quantum computing is being taught in universities the world over and there's a very rich and energetic environment of venture capital as well and major companies. So I mean when you are you're talking about your roadmap for building these devices with more and more cubits Like I don't really understand what that means. Other than its more cubits. More cubits sounds better, sounds more powerful. But when you go from 127 Cubits processing power to 4000 express like how that makes a difference or you know does it just is it gonna be like a moore's law of cubits where we always need more and more cubits on you know or is there like a sort of threshold you need to meet to kind of do the things you need to do to get that quantum advantage. Yeah good question because there is an exponential underneath it. So the number, remember I told you that the number of states that you'll get to represent, you get to represent information more richly than with just zeros and ones. So to be precise the number of states you get to represent is two to the end where n is the number of Cubans. So that's an exponential right there and just a good number of states. So just translate that. So like if you're if you're analyzing a problem instead of two things at once, you can analyze infinite number based on how many cubits you have to process that. This scenario. That's right. So basically how an algorithm works in a quantum computer and it's gonna sound a little bit abstract for a second but just bear with me for a minute. So imagine for um for a second that you have a fear right in So picture of fear and in classical world you can put a zero in the north pole of your sphere or a one in the south pole of that fear. Now in quantum world the first thing you use is you put the computer in a superposition of states and what that means is that imagine that if you put it in the equator now you would have a linear superposition of zero and one. But the reality is you can put them anywhere in your sphere. So you have all these potential combinations of representing information. So imagine that you have in that fear is populated with like a huge number of those dots in the surface of this sphere, each one of those is represented by an amplitude and a face like those little dots on that sphere like moons that can rotate. So what the next thing you do in an algorithm is you put data into the computer in practice. What it means is that you take those little moons that are in that here and you rotate it slightly different. Now your machine has all the possible states that they can represent based on the number of cubits data is encoded in the state of that machine. And now the magical act of quantum computers is that you get to actually interfere information like waves in the ocean. So the right answer is maximized. And other things you don't care about cancels out notice is so different than what we learn about having zeros and ones and transistors or switches. And you use boolean algebra to be able to do calculation. Okay, I want to stick with that for a second. I do want to remind you if you have a question, raise your hand will get you a paddle. And uh So we'll get some questions in a second. But like all right. So you just described beautifully the wave the crescendo of the wave of all the possible solutions. But like can you translate that to you said people are running these every day in large numbers. What's an example of one challenge that you solve with the quantum and how you'd reach the solution differently than if you were running a super powerful regular computer. Yeah, just Okay, so, good question. So, so they're working there was the original idea around the possibilities of why build a quantum computer, Why why do this? Was to actually model nature better? So nature? Uh, the environment around us obeys the laws of quantum mechanics. So if you actually want to understand the properties and how you know, physics operates and materials and chemistry, let's say you want to build a better battery technology uh, for electrification or let's say you want a better fertilizer for agriculture or a better catalyst for industrial process or a better alloy. Right. If you're an aerospace, that's a process of trial and error in materials development And with the advent of computers, we had the opportunity to calculate things. However, the calculations that you perform with traditional computers, the characteristic that they have is that there's an exponential number in their meaning, as you try to say, model two molecules to build a better lithium chemistry as you start looking at the interactions, the electron orbital's of the atoms are present in there. The number of calculations that you have to solve classically grows exponentially. So the reality is that even with the best supercomputers, the best we can do is to approximate an answer in quantum you don't have that problem. So the idea is to build machines that behave quantum mechanically to be able to model these problems. So for all the industrial sectors that rely on advances in materials in chemistry and physics around that this will have a huge impact. Now, interestingly enough, what happened is that clever people discover other problems that had nothing to do with modeling nature. And that also delivered you an exponential advantage compared to classically. And the most famous example of that has to do with factoring. And the reason that that matters a great deal is because that the fact that it is hard to factor numbers in classical computers is the basis of most if not all encryption that we rely to secure the modern digital economy. So now we have these two vectors of everything that has to do about modeling our world, the natural world and now this ram of like mathematics that also is going to have big implications for industry and for society. Okay, I'm seeing hands which is good. We're gonna we're gonna see what questions people have will go right here first. Could you just stand just give us your name and company franklin. Hey daria, how are you with Marsh a quick question. So I think you just answered it though. So classic computing. Quantum computing you're not saying that quantum computing will replace all classics. I thought certain datasets lend themselves more to quantum computing solutions than classic computing. But can you confirm that or? Absolutely right. So quantum computers will not replace all classical computers by any stretch of the imagination. The future is classical and quantum working in concert. In fact to run a quantum computer, you need classical computers uh to be able to do that. And it is indeed the case that for example, for lots and lots of applications running your spreadsheet or you know your word document or playing a video or so on, you're not gonna need quantum computers. There will be a subset of problems that but but the interesting thing about quantum is not that they're slightly better than a classical version. There are classes of problems that we will never be able to solve. Exactly. Never with classical computers. That's what's so exciting is because the only technology that alters the equation between what's possible and impossible to solve, but it's all about the combination. In fact I have this little sort of pseudo equation that I like to share with people that the future of computing is bits plus neurons plus cubits. And what I mean by that it's a form of high precision calculations that we do with a lot of our machines. The neural networks are the basis of artificial intelligence and the sort of cubic technology based on quantum. And I say that to emphasize the pluses each vector is fascinating. But I think what the world has been yet fully grasped is the implications of the convergence of these technologies. So the future of workloads will be you to take your problem and behind the scenes we will partition the problem to figure. What do you run with high precision calculations? What do you run with neural networks and what do you run with quantum machines and then bring it back sort of in a transparent fashion to the user? We have a question over here. Hey, my name is jake, I'm with Bushel from Fargo North Dakota. Thanks for having us here so on. Would you say on the Ai side, would you argue that quantum computing capabilities have changed the assumption around ai, could you argue that today that not just neural networks but the next level of because we know that neural networks are interesting but they're still solving for approximation sort of outcomes. Would you say this is likely the path or has this created a much more likely outcome where we'll have true human AI capabilities in quantum computing. Would would that be an argument? Well, what what I would say is that for many decades we were in um sort of intellectual and computational straitjacket, which was you know, very precise forms of like microprocessor architecture is a particular way that there was a uniformity in computing with ai over the last decade has demonstrated, is that specialized forms of computing can deliver tremendous value for very specific kind of mathematics. So in implementing neural networks and they multiply accumulate function that is inherent in there. You can have Gpu s as an example, there are much more efficient compared to a traditional CPU, there's a lot of excitement and it's a new field but it's a research field around the possibilities about this intersection between quantum and machine learning. So it's a new area, right? It's a new scientific discipline and so on. And it will require not just to cast the same problem as we've done with Gpu. S with machine learning, but to reformulate the possibilities of how do you do neural networks and how do you do training using quantum machines? So one of the most important things about to be successful in the quantum world is not to be in the mode of I run something today. This is my software today. My algorithms today. I just want the quantum machine to run it faster. That's actually a recipe for going down the wrong track. It's about giving this new way to represent information. What should I think of differently? How should I explore new possibilities of performing computation calculations in novel ways? So it requires a lot more creativity to be able to do that. But the application of quantum machine learning is one of the most exciting areas in terms of a research endeavor. I think we have one more question here in the back. Sure, My name is Blair Lacorte. I'm from a company called Ai it's actually a E Y E. But ai every time you know we build really sophisticated sensor networks. Used to be targeting systems and I. S. R. Systems now our cars need to virtualize the environment. It actually occurred to me maybe this is a bad assumption but we model things all the time whether we get you know, RGB um you know vector analysis and orthogonal data because we have to model the world almost like creating virtual reality for a car to see. But wouldn't this kind of technology be better for modeling the metaverse? I mean for modeling environments that are very sophisticated that you want to replicate quantum physics in? Okay. Sexier, if you attach metaverse to it. Yeah. I try not to have like a bingo of tech books you know. But um well you said something different so I I don't know about like the generic metaverse story but like if you were trying to model environments that behave you know quantum mechanically or behave like you know, according to the laws of physics in a way that you needed to have that level of fidelity you could imagine down that stack of doing that modeling of a combination of classical plus quantum where there will be an element of quantum computing coming uh to um you know to to help with that. So in some ways it's gonna be a little bit behind the scenes, the way, sort of like we are going to do the algorithmic primitive calculations and so on, but that doesn't mean it's gonna be, it's not going to be incredibly important. Um And what I mean by that is behind the software of the future, I'm absolutely convinced we will always have behind the scenes this element of ai behind the scenes, this element of sort of very deterministic high precision calculations. And there will be quantum algorithms behind the scenes as well. And and I think that that's that's the world that we're trying to create. But you know, I sort of close with the thought of this is actually the most exciting time in computing in many decades. There's nothing equivalent to what is happening right now than the equivalent of what happened in the 1940s and 1950s. With the advent of those like beautiful computers that you would see in bloodshed park and places like that, where people were imagine the possibilities around this. And I think it's a moment that calls for a lot of like creativity and enthusiasm around imagining possibilities. And it's the reason why there's so many institutions right now embarking on this journey, even if you tell them, hey, I cannot solve a problem better right now, but they understand that this is gonna be possible this decade and that's why the field is advancing so rapidly Dario thanks for sharing your enthusiasm for expanding our minds And hopefully now everybody can raise their hand and say they understand quantum computing. Thank you. Thank you, thank you.

GPU

GPU stock chart

Videos