Written by Kevin Jennings
In the 1960s, a brilliant scientist predicted that within 100 years computers would be twice as powerful, 1,000 times larger, and so expensive that only the five richest kings of Europe could own them. Of course that scientist was Professor Frink from The Simpsons and the prediction was a flashback written in 1996, so it’s no surprise that his thoughts on the futures of computers were hilariously wrong.
While the means of describing a computer’s power or speed have changed over time, the most common measure of computing performance for supercomputers is FLOPS, or floating point operations per second. The world’s first supercomputer was the Cray CDC 6600. At the time, it was three times faster than the previous record holder with a performance of three megaflops.
That may sound like a lot of operations per second, especially as floating point operations are complex operations. However, an average modern processor will have maximum performance of over one teraflop, and high end graphics cards will perform orders of magnitude more operations than the computer’s CPU.
This shouldn’t be a huge shock. The simplified version of Moore’s Law is that technology doubles roughly every 18 months. With the Cray 6600 having been built in 1964, technology would have doubled 32 times since then. Two to the power of 32 is over 4 billion, so it’s no surprise that today’s average computers are over a million times more powerful than the 1964 supercomputer. But that’s just an average computer, what about today’s most powerful computers?
Started in 1993, the Top500 project ranks the world’s 500 most powerful non-distributed computer systems. As the name suggests, a distributed system utilizes multiple different machines connected over a network to work towards a singular goal. Because the goal is to track the most powerful supercomputers in the world, it makes sense to exclude anything that isn’t a single machine.
The list is updated twice each year, once in June and once in November. Each new list is released to coincide with an international supercomputer conference, and its goal is to provide a reliable means to detect and track trends in supercomputing. Each Top500 list includes the name and model of the computer, its manufacturer, processing speeds, year of creation, and the site where the computer is located. It also tracks the operating systems, but unsurprisingly the computers are almost all using some variety of Linux.
Of the top 500 most powerful supercomputers in the world, 149 of them are located in the United States. That’s an impressive feat since Japan is next in line with only 32 of the world’s top supercomputers, but not nearly as impressive as China who tops the list with 173.
With it being June at the time of recording, the newest Top500 list has just come out so we can give you the most up to date rankings. And it’s a good thing, because as of this month, there’s a new champion.
Summit is a supercomputer developed by IBM for use at Oak Ridge Leadership Computing Facility, a facility at the Oak Ridge National Laboratory. It takes up an estimated 9400 square feet (873 square meters) and was the end result of a $325 contract from the United States Department of Energy. However, not all of that money went towards Summit. The contract was actually for a pair of computers, the other being named Sierra.
Sierra is housed at the Lawrence Livermore National Laboratory in California. It is operated by the National Nuclear Security Administration and is used to run nuclear weapons simulations, so there’s not really a whole lot more we’re going to be able to say about it. Besides, Sierra was the less powerful of the two computers anyway.
When Summit was completed in 2018, it stole the first place spot on the Top500 from China’s Sunway TaihuLight. The Chinese supercomputer had maintained its dominance for two years before being overtaken by Summit’s 50% increase in performance. Summit performs at 148 petaflops (1 petaflop equaling 1,000 teraflops).
With the Nuclear Security Administration having received their nuclear weapons simulator, Oak Ridge’s expensive new toy was tasked with civilian scientific research. Summit provides scientists and researchers the opportunity to solve complex tasks in the fields of energy, artificial intelligence, human health, and other research areas. It has been used for earthquake simulations, extreme weather simulations, material science, genomics, and for predicting the lifetime of neutrinos.
There was actually a third computer that was part of the Department of Energy’s plan, although this contract went to Intel and Cray. Named Aurora, the original computer was intended to release at approximately the same time as Summit and have similar performance.
However, delays combined with a change in the project’s scope have led to it not yet being complete. Aurora is planned to be completed later this year and installed at the Argonne National Laboratory in Illinois. Its expected cost is $500 million, and it is projected to have twice the performance of Frontier, a supercomputer that we will discuss shortly. Aurora’s planned functions include research on low carbon technologies, subatomic particles, cancer and cosmology.
Assembly of Fugaku began in December of 2019, and by May 13, 2020 it was finally up and running. With a performance of 422 petaflops, Fugaku had nearly tripled the performance of Summit. It was an absolute beast in terms of speed, and it surpassed the performance of the next four top supercomputers combined.
Fugaku is housed at the Riken Center for Computational Science in Kobe, Japan, and was produced by Fujitsu Limited. Fugaku is an alternate name for Mount Fuji. After Riken announced the supercomputer’s name in mid-2019, they also unveiled its official logo which depicts Mount Fuji as a symbol of “Fugaku’s high performance” and “the wide range of its users.” According to the Nikkei in 2018, the total cost of developing the supercomputer would be approximately 130 billion yen, or $1 billion.
The statement about the wide range of its users wasn’t just corporate puffery, either. Fugaku has been used to forecast weather and predict tsunamis in real time. It has been used to evaluate fuel economy and flight speeds for aircraft. It was even used to simulate the spread of Covid-19 and run research on Covid facemasks.
While the Top500 is the list we have been focusing on for today’s episode, there are other means of measuring computer performance by testing different kinds of workloads. Not only did Fugaku rank number 1 in the Top500, it also claimed the top spot in the Graph500, HPL-AI, and HPCG benchmark. It was the first supercomputer to ever lead all four rankings at the same time.
In November of 2020, Fugaku would receive an upgrade, increasing the number of processors. No other supercomputer was close to matching its performance, but there’s no harm in trying to stay on top. And it did stay on top for two full years, until finally being dethroned in June 2022.
Finally we end our list with the newest and fastest supercomputer in the world, and the world’s first exascale supercomputer. Exascale means that the computer has a full exaflop (1000 petaflops) of performance, or in this case 1.102 exaflops. It has a measured peak of 1.685 exaflops and a theoretical peak of 2 exaflops, making it ten times more powerful than its predecessor, Summit, and more than twice as powerful as the previous record holder. Just in case the prefixes are getting a little confusing, one exaflop is one quintillion floating point operations per second; surely that will make it easier to visualize.
If Fugaku was a beast, then Frontier is the entire pack. It surpasses the combined performance of the next eight most powerful supercomputers.
Like Summit, Frontier is hosted at the Oak Ridge Leadership Computing Facility in Tennessee. The computer cost $600 million to build and takes up 7,300 square feet (680 square meters). It was manufactured by Hewlett Packard, or more specifically their subsidiary Cray. That’s right, the makers of the world’s first supercomputer nearly 60 years ago have once again claimed the title with the HPE Cray EX235a.
Frontier is intended for scientific research, with the official website stating, “Frontier users will model the entire lifespan of a nuclear reactor, uncover disease genetics, and build on recent developments in science and technology to further integrate artificial intelligence with data analytics and modeling and simulation.” With the supercomputer just becoming operational, we will have to wait to see what incredible feats it will perform.
But Frontier is also topping more charts than just the Top500. It took the number one spot on the Green500, a similar biannual list that measures the most energy efficient supercomputers. It may have ten times the processing capabilities of Summit, but it uses less than two times the energy. When rogue artificial intelligence finally goes insane and eradicates humanity, it will be nice to know they were designed with the planet’s survival in mind.
Frontier wasn’t the only EX235a to debut in 2022. Hewlett Packard is providing the components to the CSC Data Center in Kajaani, Finland. The computer, named LUMI, isn’t even completed yet, and it has already debuted at number 3 on the Top500. However, even once completed it is not projected to take the number 2 slot, let alone number 1.
The project is expected to cost €144.5 million ($151 million), and is co-funded by the EuroHPC Joint Undertaking and the LUMI Consortium, which consists of ten different European nations. LUMI may be powerful, but it has different goals than raw processing capabilities. LUMI is to be powered using 100% hydroelectric energy. Rather than expensive and complex cooling systems to deal with the immense heat generated by a supercomputer, LUMI’s heat is going to be captured and used to heat nearby buildings. Frontier may have ranked at the top of the Green500 when it launched, but it seems that LUMI is destined to steal that position once it is fully operational.
Dojo is a supercomputer designed by Tesla to help handle the massive amount of video data generated by their self driving cars. The data is fed to a neural network that will use all of that information to improve their cars’ driving ability. There is a lot of hype surrounding Dojo, much of it from Elon Musk himself.
Separating the hype from the reality can be quite difficult. There are countless news stories that are constantly recirculated around social media making incredible claims, such as how Dojo will revolutionize artificial intelligence and that it is the fastest supercomputer in the world, with a potential performance of up to 1.8 exaflops. Because these articles are so prevalent, we would be remiss not to give Dojo a mention.
Most of the concrete facts we know about Dojo came from Tesla’s AI Day in 2021 when they revealed Dojo to the world. Or at least, they revealed their proprietary processing chip and one tile of what will eventually become the supercomputer. Dojo unfortunately failed to make the Top500 because it isn’t actually built yet.
The computer is supposed to be completed this year and may be revealed at Tesla’s upcoming AI Day, but the grandiose claims have cooled down a bit. Based on estimates regarding Dojo’s actual capabilities, it will probably make the top 10 or even top 5, but, barring some revolutionary development, it will not be the most powerful computer in the world.
It is believed that the Dojo is capable of 1.1 exaflops of AI compute, but that is very different from regular exaflops using the standard LINPACK benchmark system. So what does all of this mean in terms of the actual capabilities of Dojo? There’s a good chance we’ll find out later this year when it is finally unveiled.
We’ll have to wait to see how Dojo performs on benchmark tests, assuming Tesla even chooses to release that actual data. Elon Musk may have claimed that Dojo will perform at 1.8 exaflops, but Elon Musk says a lot of things.
We have continued to see the performance of supercomputers double or triple on a nearly annual basis, and there’s no reason to believe this will stop anytime soon. Until technology has advance to a point where we can upload human consciousness to the cloud, there will always be more work to be done.
It is just important to remember that the computers on the Top500 are only the computers we know about. Despite already leading the world in quantity of supercomputers, there have been reports for over a year that China secretly developed supercomputers using their own technology. Frontier may officially be the first exascale supercomputer, but it is possible that China won that race by over a year and has been operating the computer in secret to avoid U.S. sanctions. As of May of this year, China is allegedly home to two exascale computers, but there is no way to substantiate these reports.