5 minute read
Meet Mira: Argonne Lab's Supercomputer
Meet Mira
Argonne Lab'sSupercomputer
BY DICK MORTON
Huge simulations, self-teaching artificial intelligence, and analyzation of vast amounts of data are the major components for furthering human knowledge and technology. None of these daunting tasks could be possible with just any computer. In fact, it takes a computer the size of an office to do this kind of work. Much as you imagine the old computers of the ‘50’s and ‘60’s looked like, supercomputers today on the outside look similar in size and shape to their ancestors.
Since the dawn of the Information Revolution, advancements in computer technology have advanced exponentially. Today’s computer technology is proof of that with many home devices having the same computing power as a supercomputer made fifteen years ago. Home video game consoles, such as the XBox One X and the PlayStation 4 Pro have about the same computing power as Argonne labs first supercomputer, a 5-teraflop IBM Blue Gene/L prototype.
The success for this prototype supercomputer resulted in the creation of the Argonne Leadership Computing Facility (ALCF). The Facility is housed in a giant cement building with architecture reminiscent of the Guggenheim. The inside of the building is an open design where on the first floor a huge stone Zen garden is visible from any floor above. The building is growing, though, to make room for the next generation of supercomputer. Currently, ALCF has two supercomputers with one on the way.
-JINI RAMPRAKASH
Inside a very large room inside ALCF sits these mammoth supercomputers. The room is loud with ventilation that’s cooling the supercomputers and other supporting equipment. Under the floor tiles that are removable, a complex array of copper piping carries cooling water to the supercomputers from the cooling station that is on the other side of a long cement wall.
Argonne computer scientist Jini Ramprakash, Deputy Division Director of ALCF, guided HM though the tour of the facility.
“We have a lot of interaction between Nano Research Materials and Argonne’s Photon Source and our supercomputers, because as you can imagine the people come to us for this have a lot of data that they need to make sense of, or they need to process. So, they need a lot of computational power,” stated Jini, “For some people it is sufficient enough to have a normal desktop computer in front of them. Sometimes a desktop is not enough. That’s when Argonne’s supercomputers are needed.”
Meet Mira.
A 10-petaflop IBM Blue Gene/Q Supercomputer especially designed by IBM for Argonne Labs. 48 racks, 786,432 processors, and 768 terabytes of diskspace made this the third most powerful supercomputer in the world in 2013. Since Mira’s inception in 2011, this supercomputer has done many amazing things such as mapping the entire human brain and simulating the process of nuclear fission down at the atomic scale.
For Mira’s last task before hitting retirement it is simulating the entire universe through the entirety of existence! This extremely challenging task will take a total of 800 million core-hour which works out to be about six weeks of work for the supercomputer. It started on August 20, 2019 by modeling how the universe looked 50 million years after the Big Bang. Mira will then simulate billions of years to present time, and create a high-resolution model that researchers can then compare to the actual universe. This could help prove many cosmological theories and better broaden our understanding of the future of the universe.
Running a supercomputer is very expensive. The challenging cost of powering, cooling, and having a team of expert technicians and researchers are what spell out Mira’s demise after it is retired. At which time, all the copper and precious metals will be scrapped. Mira will be completely dismantled to make way for Argonne’s next supercomputer, Aurora. During the transition period the slack will be taken up by Argonne’s other supercomputer.
Meet Theta
This beast, built in 2017, consists of 24 cabinets and 4,392 compute nodes. Each computer node has a 64-core processor, 16 GB of DRAM, and 192 GB of DDR4 RAM which gives this supercomputer’s peak performance at 11.69-petaflops. Theta was designed in collaboration with Intel and Cray for Argonne Labs. Theta is a Cray XC40 system with a second-generation Intel Xeon Phi processor and the Cray Aries proprietary interconnect.
Use of these supercomputers are free of charge. Yet, it cost millions of dollars to run this facility, and all of that cost is handled by government funding. The basic agreement is if a person or a company wants to use Argonne’s supercomputers for free, they must publish all or part of their data to the public. This allows for true freedom of information and collaboration with millions of bright minds around the world. Because of this freedom of information, Argonne does not compute sensitive data or defense information. It’s purely for an open, sharing scientific community.
Argonne basically started cloud computing with their five-dimensional network. Users of Mira and Theta never actually get near the supercomputers let alone step foot into Argonne Labs. All work is requested though an online interface that connects the user’s computational data to the supercomputers. Now, Argonne Labs is making way for the next generation of supercomputers, named Aurora. For this new supercomputer the cooling system needs to be expanded. Even the room that it will share with Theta will be expanded. Expected to be finished by 2021, Aurora will be another innovative collaboration with Intel and Cray that will be capable of providing over 1,000-petaflops! This will open the door to deep learning, machine learning, and other artificial intelligence techniques. ■
Understanding FLOPS
With today’s current computers being able to multitask computations over multiple cores a much useful calculation for measuring a computers performance is needed. This newest form of measure is called FLOPS which stands for floating point operations per second. Floating point arithmetic is needed for large numbers and compilations that require a very dynamic range. Most store-bought computers can handle at least one teraflop which is one trillion floating-point operations per second. A petaflop is equal to one thousand teraflops. So, Mira is roughly 10,000 iPad Pros combined!