The term “exascale” sounds like a science fiction term, but its definition is simple and highly non-fictional. The human brain allows him to perform about one simple mathematical operation per second, while an exascale computer allows him to perform at least 10 quadrillion calculations in the time it takes. To say “One Mississippi”.
In 2022, Frontier, the world’s first declared exascale computer, will come online at Oak Ridge National Laboratory. This makes him 2.5 times faster than his second fastest computer in the world. However, it will soon become more competitive (or peer) with upcoming testers such as El Capitan, housed at Lawrence Livermore National Laboratory, and Aurora, housed at Argonne National Laboratory.
It is no coincidence that all of these machines are located in facilities with names ending with the words “National Laboratories”. The new computer is a project of the Department of Energy and its National Nuclear Security Agency (NNSA). DOE oversees these laboratories and a network of others throughout the country. NNSA is tasked with overseeing nuclear weapons stockpiles, and part of the raison d’etre of exascale computing is to perform computations that help maintain nuclear weapons stockpiles. But supercomputers also exist to solve the hardest problems of pure science.
When scientists finish Frontier’s commission dedicated to such basic research, they will learn how energy is produced, how elements are made, and how the darker parts of the universe drive evolution. We want to uncover the core truths in various fields, such as whether —all through near lifelike simulations in a way that wouldn’t have been possible without having to sniff the supercomputer of a few years ago.
“In principle, the community could have developed and deployed an exascale supercomputer much sooner, but by our standards it was not easy, convenient, or affordable,” said Oakridge. said Douglas Kothe, Deputy Laboratory Director of Computing and Computational Sciences at . Impediments such as massive parallelism, exa-energy consumption, reliability, memory, storage, and lack of software to start running on such supercomputers have hampered these standards. Years of intensive work with the high-performance computing industry has lowered these barriers and finally satisfied scientists.
Frontier can process 7x faster and hold 4x more information in memory than its predecessor. It consists of about 10,000 CPUs (or Central Processing Units) (which execute computer instructions and are usually made up of integrated circuits) and about 38,000 GPUs (Graphics Processing Units). GPUs were created to display visual content quickly and smoothly in games. However, they are reused in scientific calculations. One reason is that they are good at processing information in parallel.
Inside the Frontier, two types of processors work together. GPUs perform repetitive algebraic operations in parallel. “This frees up the CPU to direct tasks faster and more efficiently,” says Kothe. “It’s a game played in supercomputing heaven.” In Frontier, he divides a scientific problem into over a billion small pieces so that the processor can eat a small portion of each problem. I can. Next, Kothe said: You can liken each CPU to a factory crew his chief and the GPU to a frontline worker. “
The 9,472 different nodes in the supercomputer (each not really a supercomputer at all) are also all connected in such a way that information can be passed quickly from one place to another. But importantly, Frontier isn’t just faster than your old machine. It also has more memory, so you can run larger simulations and keep a lot of information in the same place that processes those data. It’s like keeping all your acrylics handy while you’re trying to do a paint-by-numbers project instead of fetching each color as needed from the other side of the table.
With such power, the Frontier and the beasts that follow could teach humans about a world that may have previously been opaque. . Chemistry lets you experiment with different molecular configurations to see which ones make better superconductors or pharmaceutical compounds. And in medicine, they’ve already analyzed all the genetic mutations in SARS-CoV-2, the virus that causes COVID, and he’s cut the time it takes to do the calculations from a week to a day, and these tweaks are We understand how it affects the transmissibility of viruses. That saved time allows scientists to perform lightning-fast iterations, change ideas, and run new digital experiments in rapid succession.
With this level of computational power, Kothe says, scientists won’t have to do the same rough calculations as they used to. For older computers, he often had to say: Maybe I don’t need that equation. In physics terms, this is called creating a “spherical cow”. It turns a complex phenomenon like a cow into a very simplified one like a sphere. Scientists hope to use Exascale his computer to avoid these shortcuts and essentially simulate cows as cows.
Frontier’s upgraded hardware is a major factor in that improvement. But hardware alone won’t do much if scientists don’t have the software to harness the new powers of the machine. That’s why an initiative called the Exascale Computing Project (ECP), which brings together the Department of Energy and its National Nuclear Security Administration, and industry partners, is sponsoring 24 early science coding projects in parallel with the development of the supercomputer. doing.
These software initiatives not only use old code to simulate a sudden onset of bad weather, for example, but they insert it into Frontier and say, “It’s not almost lightning fast, it’s lightning fast.” I made a prediction! A set of hardened and optimized code is required to get more accurate results. Kothe, who is also director of ECP, said:
But Salman Habib, head of an early science project called ExaSky, says getting the bigger answer won’t be easy. “Supercomputers are inherently brute force tools,” he says. “So we have to use them in a sensible way. I’m scratching my head and thinking, ‘How can I actually use this instrument, which might be a blunt instrument, to do what I really want to do?’ It’s fun to explore the mysterious composition of the universe and the formation and evolution of its structure. The simulation models the effects of dark matter and dark energy, and includes initial conditions to explore how the universe expanded immediately after the Big Bang.
Large-scale astronomical surveys (such as the Dark Energy Spectrometer in Arizona) have helped illuminate these shadowy corners of the universe, showing how galaxies formed, shaped and spread as the universe expanded. is shown. But the data from these telescopes alone cannot explain it. why what they see.
However, theory and modeling approaches such as ExaSky may be able to do so. If a theorist suspects that dark energy exhibits a particular behavior or that the notion of gravity is misaligned, the simulation can be tweaked to include those notions. It then spits out the digital universe, allowing astronomers to see if it matches or disagrees with what the telescope’s sensors pick up. “The role of the computer is to become the virtual universe for theorists and modelers,” he says Habib.
ExaSky expands on algorithms and software written for low-end supercomputers, but simulations have yet to lead to major breakthroughs in the nature of the dark component of the universe. The work scientists have done so far offers “an interesting combination of things that we can model, but not really understand.” However, exascale computers allow astronomer Khabib to use more cattle-like physics to simulate larger spaces at higher resolutions. Perhaps understanding is progressing.
Another early frontier project called ExaStar, led by Daniel Koesen of Lawrence Berkeley National Laboratory, explores another cosmic mystery. This effort simulates a supernova. A supernova is the end-of-life explosion of a massive star that produces heavy elements at the limit. Scientists have a rough idea of how supernovae unfold, but we don’t really know a whole cow version of these explosions or how heavy elements are created within them. No one is there.
“Supernovae and stellar explosions are fascinating events in their own right,” says Kasen. “But they’re also key figures in the cosmic story.” Their extreme reactions cannot be fully replicated in physical experiments, but digital testing is possible and non-destructive.
A third early project explores the familiar phenomenon of nuclear reactors and their reactions. The ExaSMR project uses exascale computing to figure out what’s going on under the shield of a ‘miniature modular reactor’. Early supercomputers could only model her one component of the reactor at a time. I was then able to model the entire machine, but only at one point, such as getting an accurate picture of when it was first turned on. “We are now modeling the evolution of the reactor from start-up through the fuel cycle,” says Stephen Hamilton of Oakridge, who co-leads the effort.
Hamilton’s team will investigate how neutrons move around and affect the chain reaction of nuclear fission, and how heat from fission moves through the system. In the past, understanding heat flow in both spatial and temporal detail was simply not possible because computers did not have enough memory to compute an entire simulation at once. “The next focus is on looking at a broader class of reactor designs,” Hamilton said, to improve efficiency and safety.
Of course, nuclear power has always been the flip side of that. other Use of Nuclear Reactions: Weapons. At Lawrence Livermore, Teresa Bailey leads her team of 150 people. A lot of it is busy preparing code to simulate weapons running on El Capitan. Bailey is Lawrence Livermore’s Associate Program Director in Computational Physics, overseeing part of the Advanced Simulation and Computing project (national security aspects). A team at the NNSA Institute worked on her R&D to help modernize the weapons code, with support from ECP and a more weapons-oriented effort, the Advanced Technology Development and Mitigation program.
Ask scientists if computers like Frontier, El Capitan, and Aurora are finally good enough. Researchers always need more and better analytical power. There are also external pressures to keep computing moving forward. Not just for bragging rights, which is cool, but also because better simulations could lead to new drug discoveries, new advanced materials, or new Nobel Prizes for keeping the country on top.
All these factors make scientists already talking about a “post-exascale” future. That future could include scaling exascale systems with quantum computers and more artificial intelligence. Or maybe it’s something else entirely. In fact, someone might need to run a simulation to predict the most likely outcome or most efficient path forward.
