Folding@Home Reaches Exascale: 1,500,000,000,000,000,000 Operations Per Second for COVID-19
by Anton Shilov on March 26, 2020 6:00 AM EST- Posted in
- Distributed Computing
- Exascale
- Coronavirus
- Folding@Home
Folding@home has announced that cumulative compute performance of systems participating in the project has exceeded 1.5 ExaFLOPS, or 1,500,000,000,000,000,000 floating point operations per second. The level of performance currently available from Folding@home participants is by an order of magnitude higher than that of the world’s most powerful supercomputer.
Right now, cumulative performance of active CPUs and GPUs (which have returned Work Units within the last 50 days) participating in the Folding@home project exceeds 1,5 ExaFLOPS, which is 10 times faster than performance of IBM’s Summit supercomputer benchmarked for 148.6 PetaFLOPS. To get there, Folding@Home had to employ 4.63 million CPU cores as well as nearly 430 thousand GPUs. Considering the nature of distributed computing, not all CPU cores and GPUs are online at all times, so performance available for Folding@home projects varies depending on availability of hardware.
| Folding@home Active CPUs & GPUs Reported on Wed, 25 Mar 2020 23:04:31 GMT |
||||||||
| AMD GPUs | NVIDIA GPUs | CPUs | CPU Cores | TFLOPS | x86 TFLOPS | |||
| Windows | 75,823 | 314,952 | 474,277 | 3,588,315 | 680,371 | 1,384,998 | ||
| Linux | 3,675 | 41,113 | 78,124 | 811,997 | 85,028 | 167,152 | ||
| macOS | - | - | 41,582 | 230,198 | 2,578 | 2,578 | ||
| Total | 79,498 | 356,065 | 593,983 | 4,630,510 | 767,977 | 1,554,728 | ||
| Note: | CPUs and GPUs which have returned Work Units within the last 50 days are considered Active. | |||||||
The outbreak of COVID-19 has been taxing for a number of computational biology and chemistry projects. IBM recently formed its COVID-19 High Performance Computing Consortium that pools together major supercomputers run by various research institutions and technology companies in the USA to run research simulations in epidemiology, bioinformatics, and molecular modeling. Cumulative performance of supercomputers participating in IBM’s COVID-19 HPC Consortium is 330 PetaFLOPS.
Folding@home distributed computing project uses compute capabilities to run simulations of protein dynamics in a bid to better understand them and find cures for various diseases. Recently F@H started to run projects simulating theoretically druggable protein targets from SARS-CoV-2, which attracted a lot of attention as SARS-CoV-2 and COVID-19 are clearly the hottest topics these days.
We at AnandTech also have our Folding@Home team, which are currently in a race against our sister site Tom's Hardware. If you have a GPU spare that's not too old, think about joining us in our battle. We are Team 198.
Related Reading:
- IBM & Partners to Fight COVID-19 with Supercomputers, Forms COVID-19 HPC Consortium
- Help Fight COVID-19 and Tom's Hardware: Join The Great Folding@Home Coronavirus Race
Source: Folding@Home Twitter
Facebook
Twitter
RSS
71 Comments
View All Comments
Qus - Friday, March 27, 2020 - link
This. My older pc uses only CPU because older AMD's GPU does not get any taskswebdoctors - Thursday, March 26, 2020 - link
About a week ago I started Folding at home. Hopefully my Ryzen 2700x and a Geforce 1080 GPU helped makeup half that number up there :)Does pump out a lot of heat though :(
ballsystemlord - Thursday, March 26, 2020 - link
Here's an interesting statistic:Ratio of CPUs running Windowz to Linux 6.0708233.
Ratio of CPU cores running Windowz to Linux 4.4191235.
So Linux users are more likely to buy high core count CPUs. Now if only we had a more complete break down...
BTW: Before someone tries to comment based on my bias, fanboy status, or such, I was pointing out an interesting statistic, not saying someone/thing is good or bad.
blppt - Thursday, March 26, 2020 - link
Not to nitpick, but if you don't want to be called out for bias, maybe not call Windows 'windowz' or 'windoze'.Slash3 - Friday, March 27, 2020 - link
Their math is also off.The Windows CPU results average 7.56 cores per CPU (3,588,315 / 474,277).
Linux systems average 10.39 cores per CPU (811,997 / 78,124).
MacOS systems average 5.53 cores per CPU (230,198 / 41,582).
Stragak - Friday, March 27, 2020 - link
Is it though? F@H includes Hyperthreading if your PC has it as it is a very effective means of loading a Core, they also leave one core inactive during folding so one may still perform mundane tasks.Thus it gives you 7 cores what is surprising is how few over the 'baseline' they total CPU core count is. When both of my PC's on Microsoft based OS have a combined total of 26 CPU's.
13 cores per PC? Is there even a 7 Core (if you divide out the hyperthreading) Or is it a plain Jane 4 core 4770K and a Dual 2660 Xeon.
Stragak - Friday, March 27, 2020 - link
But that is only one permutation Out of the half a million.close - Friday, March 27, 2020 - link
A 4 core HT 2 way SMT CPU is 8 cores. It's very likely that most Windows machines are running on run of the mill Intel 4 core CPUs that were *very* popular for a decade. A number of 10-12-16 core CPUs would pull the average up slightly to just over 7 cores.ajp_anton - Thursday, March 26, 2020 - link
The F@h stats report that you linked says 1,554,728 x86 TFLOPS. So is that only for the CPUs? What about all the GPUs also mentioned in the stats?Gc - Thursday, March 26, 2020 - link
According to https://foldingathome.org/support/faq/flops/ some GPUs have more powerful trancendental function units than others, so the x86 column converts each kind of op to the number of ops it would take on a x86 CPU. (I'm not clear if x86 excludes AMD64 or SIMD extensions like SSE or AVX).