Latest / Technology Now / What do you do with the world's fastest, verified supercomputer?
Transcript
- AUBREY LOVELL Hello friends and welcome back to Technology Now, a weekly show from Hewlett Packard Enterprise where we take what's happening in the world and explore how it's changing the way organizations are using technology. We’re your hosts Aubrey Lovell, MICHAEL BIRD and Michael Bird, and this week we’re following on from our previous episode and asking: what do you do with the world’s largest supercomputer? - We’ll be exploring the different programmes run on, and discoveries made on, the world’s largest supercomputers - And we’ll be looking to the future and asking what could be discovered next… AUBREY LOVELL That is correct, so if you’re the kind of person who needs to know *why* what’s going on in the world, matters to your organisation, this podcast is for you. (oh) And if you haven’t yet, subscribe to your podcast app of choice so you don’t miss out. Right, let’s get into it! MICHAEL BIRD So, as you may have picked up from the introduction, this is the second part in a two-part miniseries about the world’s largest supercomputer. It can, of course, be listened to as an independent episode but if this is the sort of thing you find interesting, make sure to check out last week’s episode “how do you build the world’s fastest supercomputer?” AUBREY LOVELL Absolutely. Now, supercomputers are used in all sorts of fields but one of their most important uses is in research and we talk about this a lot on the show right? They can analyse bigger quantities of data than a human could hope to match; they can simulate experiments which would be too dangerous or costly to perform; they can even model the explosive death of stars. Very interesting. So, Michael, quickfire question, what big question would you load into a supercomputer if you had the opportunity? MICHAEL Ooh. I think something around like how to save energy or like energy production around the world like, you know, like where are the most optimal places to put wind turbines and solar panels and like hydropower. AUBREY LOVELL Yeah, that's a good one. I was thinking kind of similar but more on a smaller scale of my day to day, right? I was thinking, okay, could I ask it to how do I make my home smarter to save energy just on my day to day and my house? And then I started getting really crazy. And like, what if it could tell me like giving me an investment portfolio and the best things to invest in in the next 20 years so I get super rich and then I can retire. But yeah, I think energy wise, absolutely. You're definitely going down the sustainability route and that's a good thing. MICHAEL BIRD I'm like, how can we save the planet? You're like, how can I make myself filthy rich? AUBREY LOVELL The balance of life… But as we mentioned last time, the world’s largest supercomputer, El Capitan, is housed at Lawrence Livermore National Laboratory and works on the U.S Nuclear Deterrent but it also has a sibling machine, Tuolumne, which works on unclassified research. MICHAEL BIRD It absolutely does, and here to tell us more about both of these systems is Rob Rieben, a computational physicist, at Lawrence Livermore National Laboratory. Rob, welcome to the show. First question: So you are a computational physicist. What does a computational physicist do? ROB RIEBEN I always describe it as kind of a hybrid between it's about 50 % physics, about 30 % mathematics, and about 20 % computer science. So it's taking all those disciplines and using it ultimately to make predictions about some complex physical system. So if we have some system that we want to predict the behavior of, we model that with physics models. So these are partial differential equations. That's basically what we're solving. But it's a lot of partial differential equations all coupled together, running at very high resolution in 3D, you know, on these big computers. MICHAEL BIRD Wow, okay, so sort of mind-melting stuff ok so How do you use a supercomputer? So as a user are you sort of sitting there with a screen and a keyboard like a normal laptop where you've just got a lot of power behind you and you can just run things a bit faster? ROB RIEBEN Yeah, in most cases, it really is just kind of sitting at either a laptop or your desktop. And you can kind of think of the supercomputer as this cloud resource or something that's just out there. And most of our scientists connect from their regular desktop or laptop. And in some cases, you could actually connect with a phone. For the most part, you're just interacting with mouse and keyboard, submitting what we call job scripts, so a simulation that a user runs is orchestrated through a job submission scheduler MICHAEL BIRD So the computer is running 24/7 and it’s… there's a queue of jobs that people want to run and it just sort of works its way through that queue. And hopefully there are constantly things being added to that queue. So it's not just sat idling. ROB RIEBEN Exactly. So when our machines come online, they very quickly become almost fully utilized. So there's usually a long line no matter how big the machine is. And in fact, for El Capitan, it's so much bigger than our previous machine. We were all kind of wondering, are we going to be able to fill this up? And lo and behold, as it's already up and running, I find myself waiting in line already, waiting to get jobs in. So there's all sorts of ways that people run many simulations so we have some people that will run one massive simulation that can take up the entire machine. But for every one of those, there's tens or hundreds or thousands of much smaller jobs where people are running what we call ensembles. So these are many, calculations to explore some space, some design space. MICHAEL BIRD So, you know, why can't they just be done on a standard desktop computer? Why do they have to be run on, I mean, in particular, the world's fastest supercomputer? What makes those problems so special? ROB RIEBEN It's fundamentally the scale. So these are physics simulations, multi-physics. So we're not talking about just fluid dynamics, for example. We're coupling in multiple physical processes. So that includes, continuum mechanics, plasma physics, radiation, hydrodynamics, lots of different physics. And we do this in a really big scale and so resolution is a common way to describe this. So if you think about a computer monitor or your television screen, that has some resolution like 4K, let's say, that's a bunch of pixels. And so you can think of a simulation as consisting of a bunch of these pixels, but we're not just limited to two dimensions here. We have very high dimensional calculations. So think of dimensions beyond two. So a three dimensional simulation, you'd have a block of pixels, but then there's other dimensions to consider depending on the physics that's enabled. And so it's the resolution, it's the size of these simulations that really demands a supercomputer. And that means being able to run in parallel. So one big job that's chopped up into small pieces and runs in parallel on a big compute system like El Capitan. MICHAEL BIRD So these problems may have been, I don't know, worked on with previous supercomputers, but actually what you can do now is increase the resolution? I.e. get more information ROB RIEBEN That's right. You can resolve more fine structure, you get higher fidelity. You can actually observe in simulation a physical phenomena that you might not fundamentally be able to capture at a lower resolution. So you can actually make scientific discovery as you increase resolution and then later go back and confirm that with experiment. That's happened multiple times. MICHAEL BIRD Okay, so who has access to use the supercomputers at the Lawrence Livermore National Lab? Is it only staff or is it other people or institutions can get access to it? ROB RIEBEN So the El Capitan machine is limited to the NNSA Tri-Lab. So that's the Lawrence Livermore, Los Alamos, and Sandia National Laboratories. And that's largely for our strategic deterrence mission. So that's an important part of this machine. And so access is limited to staff members from there. But we have a sister machine called Tuolumne, which is on our open network which is open to a lot more people. And so that we make available to academic collaborators, to industry partners, people from all over the Department of Energy, other national facilities. And that's really a big part of our, what we call open science campaign. So it's important that we get as much input from the broader scientific community in utilizing this resource and helping us push its limits and also to contribute scientific information. MICHAEL BIRD Okay, so, when it comes to El Capitan, it’s working on the nuclear deterrent but isn’t that already well understood? Haven’t those problems already been solved? ROB RIEBEN The National Nuclear Security Administration in the United States has a mission to maintain the nuclear deterrent. And part of that involves what we call virtual testing. So we no longer test weapon systems underground. We have replaced that with something called science-based stockpile stewardship, where we do experiments that are not nuclear in nature, but we also couple that with simulation. And that requires supercomputing to do that. So that is something that we're always engaged in. We’re always applying these tools to learn more about this process and to uncover new things as we scale up in the fidelity and in the resolution of our calculations. MICHAEL BIRD So it's a much safer way of doing it basically. ROB RIEBEN Yeah, that's right AUBREY LOVELL Thanks so much Rob. I’m glad to know that Tuolumne is open access. I imagine it’s pretty exciting for scientists to have the opportunity to use a machine like that for their work and the democratization of access to this sort of computing facility is really heartening to see. AUBREY LOVELL Alright then. Now it’s time for “Today I learned”, the part of the show where we take a look at something happening in the world that we think you should know about. Michael, what have you got for us this week… MICHAEL BIRD So this is quite an exciting story from my home country. The National Health Service in England has announced that they are rolling out a liquid biopsy to test for lung and breast cancers following a successful trial of the scheme. The liquid biopsy is a type of DNA blood test which detects tiny fragments of tumour DNA in a patient’s blood and it provides rapid results – in some cases up to two weeks faster than regular biopsies. The speed of the results allows drugs and treatments specially tailored to the genetic profile of the cancer to be offered immediately. The announcement estimated that up to fifteen thousand lung cancer patients could benefit from the new test every year saving up to eleven million pounds for the NHS and this marks the first time a healthcare system has taken a “blood test first” approach to diagnosing lung cancer. It is hoped that the liquid biopsy will be expanded to testing for other types of cancer too including pancreatic and gallbladder cancer. AUBREY LOVELL Wow, that's fascinating. It’s obviously an important topic, right? And it is very sensitive and it's something that we all have to, you know, definitely put a focus on, especially with technology and health and how we can try to eradicate or... prevent, know, preventative. So I think that's a great progress standpoint. MICHAEL BIRD Yeah, and I think the National Health Service in England like one of the things I think they're really focused on is trying to save money. I think this is a really good thing. Especially being rolled out into other areas. AUBREY LOVELL Definitely. Affordable and accessible. That’s what we need, for sure. Alright then, now it’s time to return to Michael and Rob to find out a bit more about the sorts of discoveries which have been made on supercomputers. MICHAEL BIRD So have any discoveries been made as a result of that sort of testing, you know virtual testing that you've done? Have any discoveries, any unexpected discoveries come from that? ROB RIEBEN Yeah, absolutely. Every time we get a new machine and we push the limits and we push resolution and fidelity and scale, we always learn something new. MICHAEL BIRD Wow. So how fast does El Capitan return results when it is used? Like how much faster is it versus a regular computer? ROB RIEBEN we measure speed in multiple ways. So it's bigger and faster in multiple dimensions. So first is the compute node. So that's kind of the building block of the system. The compute node for this machine is anywhere from two and a half to 5x faster than our previous system. And then there's the other dimension, which is scale or size. And so El Capitan is about 20 times bigger in size or in total number of compute nodes than our previous system. So when you couple the per node speed up plus the sheer increase in size, we get a massive bump in performance relative to our previous system. So what tends to happen is as we get more more powerful computers and bigger systems, the scale of the problem we solve also increases. So there is kind of this constant, which is the ability to get an answer within a day We continue to value that unit of time. And so what's happened is the amount of work we can get done in a single day just continues to grow at an unprecedented scale. There are simulations that are much more complex that take longer than a day. So some can take up to a month. So with El Capitan, a calculation that used to take a month can now be done in a couple of days MICHAEL BIRD So it's not necessarily that you're solving a problem that you were doing on previous iterations faster, it's that you're saying you're maybe asking more of those problems, maybe a high resolution to those problems as opposed to trying to do anything faster. ROB RIEBEN It's all of those. So we're pushing on all of those fronts simultaneously. And so for us, probably the most valuable thing is throughput, the ability to just push as many calculations through as possible, because that's fundamentally how we learn. That's how we gain more physics understanding and make better predictions. MICHAEL BIRD So what sort of things do you see El Capitan being used for in the future? ROB RIEBEN Yeah, so I think one of the things it really enables is this massive ensembles of simulations. And so we have these big multi-physics simulations that can take a large portion of the machine to run. in our previous machine, that would have taken maybe, let's say, half the machine. Now we can do that on a small fraction of that. And in addition, now we can run many of them. So we can do an ensemble of simulations. That lets us explore different aspects of the physics that we're trying to understand. And I think the increase in physics fidelity that we can achieve here is it's letting us understand things that previously we would have not had the resolution to really be able to see. So once we start to get data like that, it also informs our models and it can help us to improve the underlying model themselves. So I'm looking forward to some of the discoveries we make here helping to improve the fundamental physics models themselves. MICHAEL BIRD Yeah. So if you personally could run anything on El Capitan, what would it be? ROB RIEBEN I'm lucky in that I do get to run on it quite a bit. My personal interest is in what's called magneto hydrodynamics. So that's coupling fluid mechanics with Maxwell's equations, with electromagnetics. And I'm also very interested in fusion energy systems. I'm also very interested in doing what's called inverse design. So that's this idea of being able to start with a desired outcome of an experiment and then use the computer to work backwards to find what's the best initial condition that will actually give you that. MICHAEL BIRD And why... why was that... why are those... were those sorts of problems not possible on previous iterations? ROB RIEBEN Yeah, and to some extent they were. It's really the fidelity and being able to do it with the higher fidelity physics that we can now deploy on El Capitan. MICHAEL BIRD Got it, got it. And what about Tuolumne? What sort of things are being run on it at the moment? ROB RIEBEN Right, so tuolumne is what we call our open access system that we use for open science. So it's open to people much more broadly than just the national laboratory. we get a lot of value out of this because it lets us stress test the system and apply it to new and challenging problems. The other thing that we're really interested in is coupling simulation with AI agentic workflows. So that's basically having an AI agent that can help orchestrate a simulation. So a multi-physics simulation is being run or in some sense orchestrated by a collection of AI agents that they're supervised by a human that can query the system and guide it in certain directions, but then the agents are off managing the individual simulations, collecting results, and then reporting them back to the user. MICHAEL BIRD Wow, okay. And how long has Tuolumne been running for? ROB RIEBEN Let's see, Tuolumne came online around December of 2024, if I recall. So a couple of months now. MICHAEL BIRD so what are the sorts of problems Tuolumne is going to be trying to get the answers to over the next few years? have sense of the sort of problems it's going to be tackling? ROB RIEBEN Yeah, think a lot is going to come out in the computational biology realm. There's a lot of projects that are excited to use this machine to apply machine learning techniques for drug discovery and for pharmaceuticals. There's, of course, all kinds of work going on in high energy density science for fusion, studying fusion plasmas, working with a national ignition facility. A lot of exciting science is on the horizon with this machine. AUBREY LOVELL Rob, thank you so much for spending some time with us. It sounds like the future is looking really hopeful too with potential discoveries being made. I also noticed you mentioned AI Agents which we actually covered on this podcast very recently so if anyone listening wants to know more about them, do check out our episode “What is an AI Agent?”. AUBREY LOVELL Right then, we are getting towards the end of the show which means it’s time for This Week In History. Michael, remind me of last week’s clue? MICHAEL BIRD Ok so it’s 1963 and this 26-year-old cosmonaut is about to set an off-world record… So we did correctly identify this as a story from the USSR but it’s not Yuri Gagarin… MICHAEL BIRD This is the story of the first woman in space: Valentina Tereshkova. To ensure the first woman in space was a Soviet Citizen, the Soviet government began their search for a female cosmonaut almost directly after Yuri Gagarin’s flight. Tereshkova applied for the role along with over 400 applicants of whom only five were selected for training. Upon completion of the training, Tereshkova was awarded an honorary rank in the Soviet Air Force and in June of 1963, she was launched into space. This honorary rank also meant that Tereshkova became the first civilian to fly in space. Tereshkova would spend almost three days in space, and would orbit the earth 48 times. Throughout her mission, she collected data on how her body responded to being in space, kept a flight log and, as you would expect, took a lot of photos. Her photos would later be used to identify aerosol layers within our atmosphere! Pretty Cool. AUBREY LOVELL That's amazing. What a great story. MICHAEL BIRD Ok Aubrey, what’s our clue for next week? AUBREY LOVELL OK so the clue for next week is: it’s 1954 and this glowing material is starting to… power up… MICHAEL BIRD It has to be something to do with nuclear right? Like a nuclear power plant or something- AUBREY LOVELL I think it’s a strong indicator in the words that Producer Harry has put in that we are talking about nuclear energy, for sure. MICHAEL BIRD I guess we’ll find out next week. AUBREY LOVELL We definitely will. And with that, it brings us to the end of Technology Now for this week. Thank you to our guest, Rob, And of course, to our listeners. Thank you so much for joining us. If you’ve enjoyed this episode, please do let us know – rate and review us wherever you listen to episodes and if you want to get in contact with us, send us a nifty email to technology now AT hpe.com MICHAEL BIRD Yes please do send us an email. Subject line: hey Aubrey, how’s it going? Technology Now is hosted by Aubrey Lovell and myself, Michael Bird This episode was produced by Harry Lampert and Izzie Clarke with production support from Alysha Kempson-Taylor, Beckie Bird, Paul Rosien, Alissa Mitry and Renee Edwards. AUBREY LOVELL Our social editorial team is Rebecca Wissinger, Judy-Anne Goldman and Jacqueline Green and our social media designers are Alejandra Garcia, and Ambar Maldonado. MICHAEL BIRD Technology Now is a Fresh Air Production for Hewlett Packard Enterprise. (and) we’ll see you next week. Cheers!