Advertisement

If you have an ACS member number, please enter it here so we can link this account to your membership. (optional)

ACS values your privacy. By submitting your information, you are gaining access to C&EN and subscribing to our weekly newsletter. We use the information you provide to make your reading experience better, and we will never sell your data to third party members.

ENJOY UNLIMITED ACCES TO C&EN

Computational Chemistry

Podcast: Why chemists are excited by exascale computing

Reporter Ariana Remmel shares insights from their recent cover story about a new generation of supercomputers on C&EN Uncovered

by Craig Bettenhausen , Mark Feuer DiTusa
January 31, 2023

 

A person with long, black hair, wearing a labcoat, walking into a bright room. The doorway and surrounding objects are reminiscent of computer parts, almost like the scientist is walking into a computer.
Credit: Matt Chinworth
Credit: C&EN

At Oak Ridge National Laboratory, a supercomputer named Frontier has broken the exascale computing barrier, meaning it can calculate more than a million trillion floating-point operations per second. In this episode, C&EN reporters Craig Bettenhausen and Ariana Remmel discuss how Frontier works and what that kind of power could mean for computational chemistry.

C&EN Uncovered, a new project from C&EN’s podcast, Stereo Chemistry, offers a deeper look at subjects from recent cover stories. Read Remmel’s Sept. 5, 2022, cover story about exascale computing at bit.ly/3RkPjr6.

Subscribe to Stereo Chemistry now on Apple Podcasts, Spotify, or wherever you listen to podcasts.

The following is a transcript of the episode. Interviews have been edited for length and clarity.

Craig Bettenhausen: Welcome to C&EN Uncovered. I’m Craig Bettenhausen.

C&EN Uncovered is something new we’re trying on Stereo Chemistry. In each episode, we’ll take a deeper look at a recent cover story in Chemical & Engineering News. C&EN’s savvy chemistry reporters will share striking moments from their reporting, their biggest takeaways, and what got left on the cutting-room floor.

In this episode, we’re diving into a cover story from last September about exascale computing and what it could mean for chemistry. I’m here with Ari Remmel, who wrote that article. Hi, Ari!

Ariana Remmel: Hey, Craig. Thanks for having me on the podcast.

Craig: We’ll link to that article in this episode’s show notes, or you can find it in the Sept. 5 print issue of C&EN. That’s Sept. 5, 2022. Definitely check it out, if you haven’t already.

But I think a good place to start, Ari, is for you to explain what exascale computing even is. It’s kind of a big, scary word, and it was actually new to me too before I read the article.

Ariana: Yeah, so, my role with C&EN is to make sure that our audience is up to date with developments in physical chemistry, computational chemistry, astrochemistry. And exascale computers, which is a new generation of supercomputers, actually falls really nicely into the kinds of stories that I already write about.

So, chemists have been using supercomputers for about as long as supercomputers have been available to us. These systems are able to do a tremendous number of calculations, which, their computational power is measured in a unit called floating-point operations per second, and floating-point operations are basically a fancy computational arithmetic that can handle, like, scientific notation, numbers where the decimal places kind of move around. So, a standard kind of laptop computer is going to be able to do something on the order of billions of calculations per second, which is why our laptops can handle all of those tabs. So many tabs. And it’s also why they can help to render really lovely, clear displays, images, and help us visualize molecules.

But if we’re really trying to study and simulate a complex molecular system in a computer, you need much, much more computational power. So, the kinds of computers that fall in this supercomputer realm for chemists traditionally have been able to do trillions of calculations, uh, can achieve trillions of flops, which again, is this floating-point operations per second.

Craig: It’s a great name.

Ariana: It’s, it’s definitely a fun word. So, in the US right now, the most powerful supercomputer is actually in the petaflops range, it can do more than a quadrillion calculations per second. And in the case of Summit, which is a petaflops computer at Oak Ridge National Laboratory, this computer can do 200 quadrillion floating-point operations per second. And this is just, like, an astonishing amount of power to the extent that this supercomputer is the one that was used to perform a lot of the calculations that went into kind of the now-iconic model of the SARS-CoV-2 viral envelope that, I assume that, I would bet that our audience has seen kind of all over the internet and various news publications at this point.

So, all of this is to kind of set up the fact that exascale computers, and in the US, the first new exascale computer is now called Frontier, also at Oak Ridge National Lab in eastern Tennessee—this computer is going to be four or five times more powerful than Summit. And that’s because it can do, it has broken the exaflops barrier, it can do a quintillion calculations per second. That’s 1018, just absolutely astonishing.

Craig: This question may be inside baseball, but these are some crazy words that I only sort of . . . how long did it take you to wrap your head around these quintillions and exaflops and this . . . [laughter]

Ariana: Yeah, no, you’re, that’s totally fair.

I would say that a lot of what kept me up at night with this story fact-checking-wise was trying to make sure that I had the right number of zeros associated with all of these numbers. Now, I think that a lot of these, kinds of jargon terms, for most chemists maybe don’t matter as much in our everyday kinds of calculations.

However, having an understanding of the kind of astonishing level of innovation behind the hardware and its capabilities I think is really important for chemists to understand, you know, what kinds of doors are going to open in the chemical sciences as we start to apply this new type of instrument, exascale supercomputers, to delving into some of the complexities of chemical systems that were not available to us or within our view with previous, earlier models.

Craig: Yeah. So let’s talk about that a little bit, because you mentioned that there’s the proteins and lipids and things like that of the COVID-19 particle. And you can imagine how you would need a lot of computational power to model all the bending and twisting and squishiness of a protein, but who else in the chemistry world needs something like this outside of, you know, protein, lipid people?

Ariana: Yeah, so, anytime that you are trying to understand how materials, how compounds within a chemical system are moving around, you know, these molecular dynamics models, which is what was partially used for modeling the SARS-CoV-2 virion, those are going to be really important.

Craig: So it sounds like one of the things you’re saying is that in addition to being able to model more complex systems, one of the things this is going to do is let us model systems we might already be at a model, but model a much longer, sort of, event.

Ariana: Yes. And that longer time frame has a lot of scientists very, very excited, especially in, you know, the biochemical space, because if we really want to be able to create controlled computational models of biological systems to really ask questions about, you know, what is happening in our experimental designs, which, experimental results are still incredibly important when doing computational studies, because you’ve got a benchmark to something, right? You need to know that you’re actually studying something that is real. And if it’s not, then you can make refinements. But in real life, in living systems, those reactions happen across these longer timescales that have just been out of reach because of limited computational power.

Craig: So aside from chemists, who else is competing for time on this machine?

That’s really, we don’t get exclusive access.

Ariana: No, we definitely don’t. So Frontier, which I mentioned is the first exascale computer to debut in the US, is the first of three that are expected to really go live, be accessible to the public sometime in calendar year 2023. So the other two systems, Aurora at Argonne National Laboratories in Illinois, and El Capitan at Lawrence Livermore National Laboratories in California, these two systems are also going to be on line soon. You’ll notice that all three of these labs are Department of Energy national laboratories. And there’s a huge interest in having this computational power to look for more energy-efficient catalysts, certainly in the realm of chemistry, but also being able to study turbulence, to study expansion of the universe, these other big questions in physics and energy systems that require just as much computational power. So, I know that there are a number of folks in the field of cosmology who are really excited to use these exascale computers to try to see if we can create models of the beginning of the universe and try to really get down to some of the nitty-gritty details of how the universe is expanding.

Craig: You went there, you’ve been to this facility. What was it like to be in this otherworldly sort of a place?

Ariana: So I drove from Little Rock, Arkansas, to Oak Ridge, Tennessee, which is just a little bit north of Knoxville on kind of the eastern edge of the state. And this is, like, beautiful, mountainous landscapes with these lush, green trees. It’s not technically part of the Appalachian Mountains, but it’s not terribly far from Great Smoky Mountains National Park.

And so the landscape is absolutely gorgeous. And in order to actually get to Oak Ridge, like you’ve got to . . . first of all, you have to have security clearance in order to get past the gate. So, you know, I showed up, gave them my ID. But then the road is just, the advice I had was just keep going, you’ll know when you get there, you know, and it was just absolutely beautiful being in these, this kind of mountainous valley. But Oak Ridge, the actual campus, is really beautiful, and they have, it looks a lot like a standard kind of research campus. But one of the, one of the sources that I spoke to, Bronson Messer, who’s the director of science for the Oak Ridge Leadership Computing Facility, he says, y0eah, if you look at these buildings, a lot of them were built in, like, the 1940s when the, when the lab was originally established. But the computing facility kind of looks like it’s been, like, steampunk retrofitted, because there’s all of these, like, sort of bits of, you know, duct and retrofitting that they’ve done to it to make sure that they can fit all of these computers, power them, store them, which is actually kind of wild.

So walking in there’s this really cool viewing platform where folks can actually look down into, it’s like glass windows, you look down into this big white room, and what’s directly in front of you, is the first bay of cabinets for Frontier. And it’s got “Frontier” written in nice big letters. So you know exactly what you’re looking at. And there are engineers moving in and out of these cabinets or, or moving around these cabinets. There’s 74 of them, and each of them weighs about 8,000 lb, partially because of just the sheer number of components that’s in there but also, you know, because it’s pumping 6,000 gal of this cooling water a minute. So they’re very, very heavy. And one of the things they had to do was, like, reinforce the floors in order to actually make sure that it, the, the floor could hold all of that computing power.

When I went to visit the facility in July, they were, the engineers and the facilities managers were still trying to “shake out the nodes,” right, they needed to be able. . . I mean, it’s just, it’s really, really funny to think about the fact that, like, there are just so many components in this machine that things could go wrong, right? And so they have to be able to make sure that each of these components is functioning properly.

But the thing that was actually really astonishing to me, is, you know, when components fail, they tend to fail at the beginning of their lifetime in a machine. So that’s one of the reasons they needed to make sure that they were doing so much work to check all of the different hardware bits at the beginning. When I was there, there was an engineer there at, like, a little, a little bench with a screwdriver, like pulling these components again that are just innovative, all on their own, taking them out, putting them back in. And he had a little CVS bottle of isopropanol there to help with the cleaning. And it was just kind of, like, astonishing, right, remembering, and having, like, a, like being in a room where there’s a clear connection between the people who had built and designed this machine and what is going to be, like, the incredible science that comes out of it.

Advertisement

And I think that one of the points that I talked with folks at Oak Ridge about was at the beginning I was struggling with, OK, do I call Frontier a machine or an instrument? And certainly by the standards of, like, computers are machines, it’s definitely a machine, but most of the researchers that I spoke to are really interested in how users of these supercomputers are going to use them, right? And so from that standpoint, thinking of this more as an instrument that is a tool for scientists rather than something that’s just going to be able to vomit out, self-generate scientific breakthroughs. You know, it’s really the people that are going to bring this new age of supercomputing to, to bear.

Craig: Yeah. This shaking out the nodes reminds me when I, I toured, there’s a science history museum at Harvard, and they have one of the early, I mean, it was a supercomputer for them. It’s, like, less powerful than your pocket calculator now, but it described, you know, hunting for bugs, which at that time was literally going into the machine and literal, actual insects would be stuck in there.

Ariana: Yeah, it’s absolutely wild. And I don’t, I don’t think there are any bugs in Frontier. I mean, it was, like, a wildly clean space, you know, sort of like a sci-fi-level white room, black monoliths of this computer with only a faint buzzing. It was actually fairly quiet in there, which I thought was kind of remarkable. I don’t know that there were any bugs. I did try to figure out how many screws there are, but no one, I, I wasn’t able to get a good answer out of that. And you know, this Frontier and the other DOE exascale computers that are scheduled to come on line are all part of a larger project called the Exascale Computing Project, which was established barely more than 5 years ago, right, and so the fact that Frontier was able to get installed during COVID, that it’s kept on schedule, and that it’s really exceeding expectations in so many different regards, given how short a time frame from proposal to actual execution is, is really also pretty remarkable.

Craig: So what are you looking at next in the field of these big computers?

Ariana: Yeah. I mean, I think that there are a lot of really interesting questions that I’m looking forward to following. Some of just the basics of how we can use artificial intelligence and machine learning to streamline some of the computational tasks that chemists are interested in doing. And I know that was a lot of jargon right there, but in order to teach a computer even the basics of a chemical intuition that we develop over the course of gen chem and all of those basic chemistry courses plus our time in lab, in order for these computers to learn that, they need these massive data sets to be able to start to make the connections, which is how machine learning works, right? If you think about a system like Dall-E 2, or even Craiyon, these kinds of AI image-generating tools that have gone viral on social media, like, those technologies are often developed with millions of labeled images, right? All of these different parameters. And we’re just not really at the point now where, number 1, we even have data sets of chemical systems that have enough of the parameters kind of worked in, but we’ve also not had the ability to process all of that data in a way that a machine learning algorithm could really try to start establishing the connections so that it could “think for itself” on a never-before-seen system to itself. So, having the ability to train these algorithms with larger data sets that can come from experimentation—which, you know, exascale computing is going to push the need for—but also being able to train them on the exascale computers because they can allow the machine learning to have all of the different nodes that it needs to make those connections is going to give us access to algorithms that will likely have utility in other areas of chemistry, right? So this is, I think, one of the places where I’m excited to see how exascale computing breakthroughs could have direct implications for chemists who may never actually run an experiment on Frontier or any of the other DOE supercomputers.

Craig: Ari, thank you so much for taking the time to tell us all about this.

Ariana: Yeah, thank you so much for speaking with me about the story. It’s very exciting. Please check out the full version online, and for more updates as I write them, you know, you can follow me on Twitter: @science_ari.

Craig: Once again, you can find Ari’s cover story about exascale computing on C&EN’s website or in the Sept. 5, 2022, print issue of C&EN. We’ve put a link in the show notes along with the episode credits. You can find me on social media as @CraigOfWaffles.

This has been C&EN Uncovered, a new series from C&EN’s Stereo Chemistry. Stereo Chemistry is the official podcast of Chemical & Engineering News. C&EN is an independent news outlet published by the American Chemical Society.

Thanks for listening.

CORRECTION:

The transcript was updated on Feb. 2, 2023, to accurately reflect the measure of the unit exaflops described in the episode recording. A quintillion calculations per second is 1018, not 1018.

Article:

This article has been sent to the following recipient:

1 /1 FREE ARTICLES LEFT THIS MONTH Remaining
Chemistry matters. Join us to get the news you need.