>>We are back in the final stretch at Supercomputing 22 here in Dallas. I'm your host Paul Gillum with my co-host Dave Nicholson, and we've been talking to so many smart people this week. It just, it makes, boggles my mind are next guest. J Poso is the HPC and AI technology strategist at Dell. Jay also has a PhD in astronomy from the University of Texas. And I'm guessing you were up watching the Artemis launch the other night? >>I, I wasn't. I really should have been, but, but I wasn't, I was in full super computing conference mode. So that means discussions at, you know, various venues with people into the wee hours. >>How did you make the transition from a PhD in astronomy to an HPC expert? >>It was actually really straightforward. I did theoretical astrophysics and I was modeling what white dwarfs look like when they create matter and then explode as type one A super Novi, which is a class of stars that blow up. And it's a very important class because they blow up almost exactly the same way. So if you know how bright they are physically, not just how bright they appear in the sky, but if you can determine from first principles how bright they're, then you have a standard ruler for the universe when they go off in a galaxy, you know how far the galaxy is about how faint it is. So to model these though, you had to understand equations of physics, including electron degeneracy pressure, as well as normal fluid dynamics kinds of of things. And so you were solving for an explosive burning front, ripping through something. And that required a supercomputer to have anywhere close to the fat fidelity to get a reasonable answer and, and hopefully some understanding. >>So I've always said electrons are degenerate. I've always said it and I, and I mentioned to Paul earlier, I said, finally we're gonna get a guest to consort through this whole dark energy dark matter thing for us. We'll do that after, after, after the segment. >>That's a whole different, >>So, well I guess super computing being a natural tool that you would use. What is, what do you do in your role as a strategist? >>So I'm in the product management team. I spend a lot of time talking to customers about what they want to do next. HPC customers are always trying to be maximally productive of what they've got, but always wanting to know what's coming next. Because if you think about it, we can't simulate the entire human body cell for cell on any supercomputer day. We can simulate parts of it, cell for cell or the whole body with macroscopic physics, but not at the, you know, atomic level, the entire organism. So we're always trying to build more powerful computers to solve larger problems with more fidelity and less approximations in it. And so I help people try to understand which technologies for their next system might give them the best advance in capabilities for their simulation work, their data analytics work, their AI work, et cetera. Another part of it is talking to our great technology partner ecosystem and learning about which technologies they have. Cause it feeds the first thing, right? So understanding what's coming, and Dell has a, we're very proud of our large partner ecosystem. We embrace many different partners in that with different capabilities. So understanding those helps understand what your future systems might be. That those are two of the major roles in it. Strategic customers and strategic technologies. >>So you've had four days to wander the, this massive floor here and lots of startups, lots of established companies doing interesting things. What have you seen this week that really excites you? >>So I'm gonna tell you a dirty little secret here. If you are working for someone who makes super computers, you don't get as much time to wander the floor as you would think because you get lots of meetings with people who really want to understand in an NDA way, not just in the public way that's on the floor, but what's, what are you not telling us on the floor? What's coming next? And so I've been in a large number of customer meetings as well as being on the floor. And while I can't obviously share the everything, that's a non-disclosure topic in those, some things that we're hearing a lot about, people are really concerned with power because they see the TDP on the roadmaps for all the silicon providers going way up. And so people with power comes heat as waste. And so that means cooling. >>So power and cooling has been a big topic here. Obviously accelerators are, are increasing in importance in hpc not just for AI calculations, but now also for simulation calculations. And we are very proud of the three new accelerator platforms we launched here at the show that are coming out in a quarter or so. Those are two of the big topics we've seen. You know, there's, as you walk the floor here, you see lots of interesting storage vendors. HPC community's been do doing storage the same way for roughly 20 years. But now we see a lot of interesting players in that space. We have some great things in storage now and some great things that, you know, are coming in a year or two as well. So it's, it's interesting to see that diversity of that space. And then there's always the fun, exciting topics like quantum computing. We unveiled our first hybrid classical quantum computing system here with I on Q and I can't say what the future holds in this, in this format, but I can say we believe strongly in the future of quantum computing and that this, that future will be integrated with the kind of classical computing infrastructure that we make and that will help make quantum computing more powerful downstream. >>Well, let's go down that rabbit hole because, oh boy, boy, quantum computing has been talked about for a long time. There was a lot of excitement about it four or five years ago, some of the major vendors were announcing quantum computers in the cloud. Excitement has kind of died down. We don't see a lot of activity around, no, not a lot of talk around commercial quantum computers, yet you're deep into this. How close are we to have having a true quantum computer or is it a, is it a hybrid? More >>Likely? So there are probably more than 20 and I think close to 40 companies trying different approaches to make quantum computers. So, you know, Microsoft's pursuing a topol topological approach, do a photonics based approach. I, on Q and i on trap approach. These are all different ways of trying to leverage the quantum properties of nature. We know the properties exist, we use 'em in other technologies. We know the physics, but trying the engineering is very difficult. It's very difficult. I mean, just like it was difficult at one point to split the atom. It's very difficult to build technologies that leverage quantum properties of nature in a consistent and reliable and durable way, right? So I, you know, I wouldn't wanna make a prediction, but I will tell you I'm an optimist. I believe that when a tremendous capability with, with tremendous monetary gain potential lines up with another incentive, national security engineering seems to evolve faster when those things line up, when there's plenty of investment and plenty of incentive things happen. >>So I think a lot of my, my friends in the office of the CTO at Dell Technologies, when they're really leading this effort for us, you know, they would say a few to several years probably I'm an optimist, so I believe that, you know, I, I believe that we will sell some of the solution we announced here in the next year for people that are trying to get their feet wet with quantum. And I believe we'll be selling multiple quantum hybrid classical Dell quantum computing systems multiple a year in a year or two. And then of course you hope it goes to tens and hundreds of, you know, by the end of the decade >>When people talk about, I'm talking about people writ large, super leaders in supercomputing, I would say Dell's name doesn't come up in conversations I have. What would you like them to know that they don't know? >>You know, I, I hope that's not true, but I, I, I guess I understand it. We are so good at making the products from which people make clusters that we're number one in servers, we're number one in enterprise storage. We're number one in so many areas of enterprise technology that I, I think in some ways being number one in those things detracts a little bit from a subset of the market that is a solution subset as opposed to a product subset. But, you know, depending on which analyst you talk to and how they count, we're number one or number two in the world in supercomputing revenue. We don't always do the biggest splashy systems. We do the, the frontier system at t, the HPC five system at ENI in Europe. That's the largest academic supercomputer in the world and the largest industrial super >>That's based the world on Dell. Dell >>On Dell hardware. Yep. But we, I think our vision is really that we want to help more people use HPC to solve more problems than any vendor in the world. And those problems are various scales. So we are really concerned about the more we're democratizing HPC to make it easier for more people to get in at any scale that their budget and workloads require, we're optimizing it to make sure that it's not just some parts they're getting, that they are validated to work together with maximum scalability and performance. And we have a great HPC and AI innovation lab that does this engineering work. Cuz you know, one of the myths is, oh, I can just go buy a bunch of servers from company X and a network from company Y and a storage system from company Z and then it'll all work as an equivalent cluster. Right? Not true. It'll probably work, but it won't be the highest performance, highest scalability, highest reliability. So we spend a lot of time optimizing and then we are doing things to try to advance the state of HPC as well. What our future systems look like in the second half of this decade might be very different than what they look like right. Now. >>You mentioned a great example of a limitation that we're running up against right now. You mentioned an entire human body as a, as a, as an organism >>Or any large system that you try to model at the atomic level, but it's a huge macro system, >>Right? So will we be able to reach milestones where we can get our arms entirely around something like an entire human organism with simply quantitative advances as opposed to qualitative advances? Right now, as an example, let's just, let's go down to the basics from a Dell perspective. You're in a season where microprocessor vendors are coming out with next gen stuff and those next NextGen microprocessors, GPUs and CPUs are gonna be plugged into NextGen motherboards, PCI e gen five, gen six coming faster memory, bigger memory, faster networking, whether it's NS or InfiniBand storage controllers, all bigger, better, faster, stronger. And I suspect that systems like Frontera, I don't know, but I suspect that a lot of the systems that are out there are not on necessarily what we would think of as current generation technology, but maybe they're n minus one as a practical matter. So, >>But yeah, I mean they have a lifetime, so Exactly. >>The >>Lifetime is longer than the evolution. >>That's the normal technologies. Yeah. So, so what some people miss is this is, this is the reality that when, when we move forward with the latest things that are being talked about here, it's often a two generation move for an individual, for an individual organization. Yep. >>So now some organizations will have multiple systems and they, the system's leapfrog and technology generations, even if one is their real large system, their next one might be newer technology, but smaller, the next one might be a larger one with newer technology and such. Yeah. So the, the biggest super computing sites are, are often running more than one HPC system that have been specifically designed with the latest technologies and, and designed and configured for maybe a different subset of their >>Workloads. Yeah. So, so the, the, to go back to kinda the, the core question, in your opinion, do we need that qualitative leap to something like quantum computing in order to get to the point, or is it simply a question of scale and power at the, at the, at the individual node level to get us to the point where we can in fact gain insight from a digital model of an entire human body, not just looking at a, not, not just looking at an at, at an organ. And to your point, it's not just about human body, any system that we would characterize as being chaotic today, so a weather system, whatever. Do you, are there any milestones that you're thinking of where you're like, wow, you know, I have, I, I understand everything that's going on, and I think we're, we're a year away. We're a, we're, we're a, we're a compute generation away from being able to gain insight out of systems that right now we can't simply because of scale. It's a very, very long question that I just asked you, but I think I, but hopefully, hopefully you're tracking it. What, what are your, what are your thoughts? What are these, what are these inflection points that we, that you've, in your mind? >>So I, I'll I'll start simple. Remember when we used to buy laptops and we worried about what gigahertz the clock speed was Exactly. Everybody knew the gigahertz of it, right? There's some tasks at which we're so good at making the hardware that now the primary issues are how great is the screen? How light is it, what's the battery life like, et cetera. Because for the set of applications on there, we we have enough compute power. We don't, you don't really need your laptop. Most people don't need their laptop to have twice as powerful a processor that actually rather up twice the battery life on it or whatnot, right? We make great laptops. We, we design for all of those, configure those parameters now. And what, you know, we, we see some customers want more of x, somewhat more of y but the, the general point is that the amazing progress in, in microprocessors, it's sufficient for most of the workloads at that level. Now let's go to HPC level or scientific and technical level. And when it needs hpc, if you're trying to model the orbit of the moon around the earth, you don't really need a super computer for that. You can get a highly accurate model on a, on a workstation, on a server, no problem. It won't even really make it break a sweat. >>I had to do it with a slide rule >>That, >>That >>Might make you break a sweat. Yeah. But to do it with a, you know, a single body orbiting with another body, I say orbiting around, but we both know it's really, they're, they're both ordering the center of mass. It's just that if one is much larger, it seems like one's going entirely around the other. So that's, that's not a super computing problem. What about the stars in a galaxy trying to understand how galaxies form spiral arms and how they spur star formation. Right now you're talking a hundred billion stars plus a massive amount of inter stellar medium in there. So can you solve that on that server? Absolutely not. Not even close. Can you solve it on the largest super computer in the world today? Yes and no. You can solve it with approximations on the largest super computer in the world today. But there's a lot of approximations that go into even that. >>The good news is the simulations produce things that we see through our great telescopes. So we know the approximations are sufficient to get good fidelity, but until you really are doing direct numerical simulation of every particle, right? Right. Which is impossible to do. You need a computer as big as the universe to do that. But the approximations and the science in the science as well as the known parts of the science are good enough to give fidelity. So, and answer your question, there's tremendous number of problem scales. There are problems in every field of science and study that exceed the der direct numerical simulation capabilities of systems today. And so we always want more computing power. It's not macho flops, it's real, we need it, we need exo flops and we will need zeta flops beyond that and yada flops beyond that. But an increasing number of problems will be solved as we keep working to solve problems that are farther out there. So in terms of qualitative steps, I do think technologies like quantum computing, to be clear as part of a hybrid classical quantum system, because they're really just accelerators for certain kinds of algorithms, not for general purpose algorithms. I do think advances like that are gonna be necessary to solve some of the very hardest problem. It's easy to actually formulate an optimization problem that is absolutely intractable by the larger systems in the world today, but quantum systems happen to be in theory when they're big and stable enough, great at that kind of problem. >>I, that should be understood. Quantum is not a cure all for absolutely. For the, for the shortage of computing power. It's very good for certain, certain >>Problems. And as you said at this super computing, we see some quantum, but it's a little bit quieter than I probably expected. I think we're in a period now of everybody saying, okay, there's been a lot of buzz. We know it's gonna be real, but let's calm down a little bit and figure out what the right solutions are. And I'm very proud that we offered one of those >>At the show. We, we have barely scratched the surface of what we could talk about as we get into intergalactic space, but unfortunately we only have so many minutes and, and we're out of them. Oh, >>I'm >>J Poso, HPC and AI technology strategist at Dell. Thanks for a fascinating conversation. >>Thanks for having me. Happy to do it anytime. >>We'll be back with our last interview of Supercomputing 22 in Dallas. This is Paul Gillen with Dave Nicholson. Stay with us.