Thousands of PCs break exaFLOP barrier

There’s a race to deploy the first supercomputer in the U.S. to break the exaFLOP barrier. Intel says it will be first with Aurora in 2021, while AMD and Cray claim they will be first with Frontier. Either way, the Department of Energy will win, because both computers will be deployed at DOE facilities.

An exaFLOP is one quintillion (10¹⁸) floating-point operations per second, or 1,000 petaFLOPS. To match what a one exaFLOP computer system can do in just one second, you’d have to perform one calculation every second for 31,688,765,000 years.

While the supercomputing stalwarts continue to build their systems, Folding@Home just crossed the exaFLOP barrier ahead of IBM, Intel, Nvidia, and the Department of Energy.

Folding@home is a distributed computing project running for 20 years. It was administered first by the chemistry department at Stanford University and as of last year, by Washington University in St. Louis. Its software runs on individual PCs and remains idle as long as the computer is in use, then it kicks in when the PC is idle.

The project simulates how proteins misfold and cause diseases such as cancer and Alzheimer’s Disease. Proteins self-assemble in a process called folding. When a protein misfolds, disease can occur. By simulating protein misfolds, Folding@Home seeks to understand why they misfold and perhaps how to prevent it and undo the damage. Over the course of 20 years the project has yielded 233 research papers, understandable to no one except PhD chemists and molecular biologists.

The technical challenge is that proteins fold in a millisecond, and simulating it and all of the potential variations requires massive amounts of computational cycles. That’s where Folding@Home comes in. Folding@Home basically operates on brute-force trial and error. Every computer running its software tries a variation of a protein fold down at the molecular level until it finds a result.

The project has primarily focused on diseases like cancer, Parkinson’s, HIV, and Ebola. In February, the team added the COVID-19 virus to its list of targets and began simulating the dynamics of COVID-19 proteins to help hunt for potential therapeutic options.

In the case of COVID-19, Folding@Home sought to better understand how the virus interacts with the human ACE2 receptor required for viral entry into human host cells, and how researchers might be able to disrupt their interaction through the design of new therapeutic antibodies or small molecules. In short, it’s trying to find ways to block the virus from entering human cells.

According to Dr. Greg Bowman, a WU professor who runs the program, Folding@Home went from 30,000 volunteers running the software in February to 400,000 in March, and then up to 700,000 users. There were so many users the database ran out of potential simulations for computers to crunch.

Last week, the administrators of the Folding@Home network reported that the network has passed the one exaFLOP mark, achieving peak performance of 1.5 exaFLOPs. That makes it more than seven times faster than the world’s fastest supercomputer, Summit, at the Oak Ridge National Laboratory. To put it another way, that’s more raw compute power than the top 100 supercomputers in the world, combined.

That’s not to say Summit is a chump. Far from it. The massive supercomputer was brought to bear on COVID-19, and in the span of a month it found 77 potential drug compounds to stop the virus.

In addition, Folding@Home’s raw compute power comes with a caveat. The thousands of PCs working on distributed computing projects like SETI@Home, Folding@Home,  and World Community Grid all work independently and are not linked. The results of one worker are not dependent on the results of another worker. The individual PCs don’t even know about each other.

In a high-performance computing system like Summit, all of the nodes are interlinked and communicate with each other, so the results from one node can be passed on and shared to others. Massive processing jobs on Summit, Sierra, and other supercomputers inherently rely on the results of computations. Step A is needed for Step B, and Step B is needed for Step C, and so on. Distributed computing projects can’t do that.

To join Folding@Home, download the client and install it. Don’t just go with the default settings; you might want to customize it. For example, do you want to let it run constantly or only when your PC is idle? And I can tell you that even at the medium power setting, it really takes over your computer and slows everything down. Not to mention heating up the CPU and GPU. So if you don’t have good cooling on your PC, don’t run it above the low power setting.

Another tip: if you want to run it on a work PC, ask first. You might get into trouble.