Folding@Home Crushes Exascale Barrier, Now Faster Than Dozens of Supercomputers

Credit: Tnguyen2791/CC0 1.0

Four hundred and seventy petaflops? Pfft. Small potatoes, these days. The Folding@Home biomedical distributed computing project has announced it now has over an exaflop of processing power as of March 25, after hitting 470 petaflops just four days ago. Estimates released by others have suggested it could be up to as much as 1.5 exaflops at this point, but either number would be astonishing. For comparison, the single fastest supercomputer, the IBM Summit, can sustain a bit less than 150 petaflops.

https://platform.twitter.com/widgets.js

If the network is limited to “just” one exaflop, then the Folding@Home network is now as fast as the 25 fastest supercomputers on Earth combined. If it has reached the 1.5-exaflop mark, as estimated by some members of the PCMR Folding@Home team, it’s now roughly as fast as the top 103 systems on the TOP500. The reason for the sharp disparity is that the sustained performance of supercomputers falls off pretty steeply as you move down the upper part of the list before flattening. Summit can maintain 148.6 petaflops, while the number 10 system is capable of just 18.2. The 50th sustains 4.289 petaflops, and the 100th is a 2.7 petaflop system.

Either way, this is known in scientific HPC circles as “an absolute f***ton of computing power.”

According to the Folding folks, there are so many people coming online, you may have to wait for a work unit to be assigned to you. If you notice that your system is pausing for longer and longer periods of time in-between WU requests, that’s because there’s a timer that increments between asks to avoid swamping the server. There’s a full FAQ on how to deal with various errors you can check for information here if you are having problems.

There is also another option for those who want to donate compute cycles to a worthy cause. Rosetta@Home is another protein-folding project based on BOINC that has fewer users than Folding@Home but has also been used to successfully predict the atomic-scale structure of the SARS-CoV-2 virus. Here’s how Wikipedia summarizes the difference between the two:

Folding@home almost exclusively uses all-atom molecular dynamics models to understand how and why proteins fold (or potentially misfold, and subsequently aggregate to cause diseases). In other words, Folding@home’s strength is modeling the process of protein folding, while Rosetta@home’s strength is computing protein design and predicting protein structure and docking.

Some of Rosetta@home’s results are used as the basis for some Folding@home projects. Rosetta provides the most likely structure, but it is not definite if that is the form the molecule takes or whether or not it is viable. Folding@home can then be used to verify Rosetta@home’s results, and can provide added atomic-level information, and details of how the molecule changes shape.

Rosetta sounds like a worthy project, either in its own right or as an alternative to FAH if the latter has run short of work. There is, of course, no guarantee that either Folding or Rosetta will find useful information in time to be helpful to the ongoing battle against Covid-19, but the larger these projects get, the more likely the chances that they will.

I sent Horst Simon an email to inquire whether the Folding@Home exascale achievement counts as far as his “No exascale computing in 2020” bet. I’m guessing he’ll say it doesn’t.

Image credit: Tnguyen2791 / CC0 1.0

Now Read:

Leave a Reply

Your email address will not be published. Required fields are marked *