News Folding@Home Network Breaks the ExaFLOP Barrier In Fight Against Coronavirus

[Admin]

The Folding@Home network has broken the one exaFLOP barrier as users from around the world combine powers to beat the coronavirus.

Folding@Home Network Breaks the ExaFLOP Barrier In Fight Against Coronavirus : Read more

 

[decaffeinatedMonkey]

This is purely epic. Wondering if they'll dish out other work unit types for other diseases in the meantime until they ramp up more COVID-19 units. I also hope this level of involvement continues past this pandemic.

 

[grimfox]

Wondering if they'll dish out other work unit types for other diseases in the meantime until they ramp up more COVID-19 units. I also hope this level of involvement continues past this pandemic.

My understanding is that covid-19 WU are included under the "other diseases" focus and that while you can bias your PC towards one of a couple of causes, F@H will send other WUs per availability. Given the current strain on their servers participants will get whatever WU is first in queue. This effort to focus on covid-19 will undoubtedly result in a lot of progress for research on other diseases.

I think one of the big things that had held F@H back has been public awareness. Now that the public is aware and they are making headlines like this I would hope that this enables increased momentum going forward.

 

[Unolocogringo]

Another anticle !!!!!
Long time folder here and love all the interest we can get for Folding.
Thanks for bringing folding to more peoples attention.
Have Crashman or one of the reviewers do a number on cooling tweeks. Folding is putting your computer to max load and we are getting a lot of forum posts about peoples computers overheating from folding. Just a thought.
And to everyone else
THANKS FOR FOLDING

 

[Badelhas]

I tried it today but saw now reference to covid 19 on the settings, only cancer and alzeimer.
Why is that?

 

[spentshells]

more power sure, but what about the internet connections?

 

[Unolocogringo]

I tried it today but saw now reference to covid 19 on the settings, only cancer and alzeimer.
Why is that?

They have not updated the software to select just CV-19.
Just select all diseases for now. You will get some CV-19 work units but not all will be for it.

With the huge rush of folders in the last few weeks the volunteers there have been doing everything in their power to meet demand for work units.
Things are a little better today and hopefully will improve over the next few days.

Folding @ Home is mostly run and supported by volunteers and donations. It receives some government funding. But that does not include money for servers or storage or salaries for people to program/run/repair/network etc... the computer systems we connect to.
So please Everyone, we should be giving these people a standing ovation!!!
Not bickering and complaining amongst ourselves about the lack of work units and our points going down

We hit a major computing milestone today in computing power all supported by VOLUNTEERS !!

 

[Angelman]

Thanks Toms for joining the massive pool of ignorant reporting on the "worst pandemic in modern history". You should've stopped that paragraph at "runs in the background." Who knows, maybe by modern history you meant the last 20 years? You'll still be wrong, be at least you wouldn't sound uninformed. For anyone interested in that dumb little thing that the media can't get right, facts, check out the W.H.O.'s report for infection rate and deaths (currently sitting at less than a 4.5% mortality rate) at:

https://www.who.int/docs/default-so...0325-sitrep-65-covid-19.pdf?sfvrsn=2b74edd8_2

Also, if modern history is allowed to extend 100 years ago unlike the 20 years or so used by this author, read about the Spanish flu of 1918 that killed 25 million people in 6 months with a total of an estimated 50 - 100 million, and a total of infected estimated at 500 million. That's just one and the list is a lot longer. For the easily triggered sensitives out there, this does not take away from the caution one should have in this current "pandemic". One should ALWAYS practice the cleanliness guidelines that were put out by the CDC. As far as the Folding@Home report goes, it's awesome that so many have volunteered their spare computing power to find a cure for COVID-19. I hope that the same enthusiasm is exhibited when other virus' are discovered and that more people stay online after this to try and discover cures for current diseases and ailments.

 

[cfbcfb]

Another anticle !!!!!
Long time folder here and love all the interest we can get for Folding.
Thanks for bringing folding to more peoples attention.
Have Crashman or one of the reviewers do a number on cooling tweeks. Folding is putting your computer to max load and we are getting a lot of forum posts about peoples computers overheating from folding. Just a thought.
And to everyone else
THANKS FOR FOLDING

Got a link for the cooling tricks. I've stress tested my 3700x machine to 100% cpu and 100% GPU and the temps stayed normal. With folding@home and a cpu load, the cpu was only at 83% but my cpu temps went into the mid 90c range even with the side off and a big fan blowing in. I set the affinity to just one core and it dropped to 10% cpu, but the temps stayed in the 85c range, still a bit warm. Folding must use aspects of the cpu (like integer/fp?) that normal workloads don't lean on as much. About the only thing I could do is water cool it, but that'd be a major job since its in a case you can't remove the back of, so I'd have to pull the motherboard.

My 6700K with a huge water cooler on it is keeping up <80C and my 6700 rig with a large air cooler stays around 78C, but I sure would like to get the 3700x into the mix.

Also need more work units! I had all 3 of them on all day the last few days and only caught about 8 hours of work units.

 

[PBme]

Got a link for the cooling tricks. I've stress tested my 3700x machine to 100% cpu and 100% GPU and the temps stayed normal. With folding@home and a cpu load, the cpu was only at 83% but my cpu temps went into the mid 90c range even with the side off and a big fan blowing in. I set the affinity to just one core and it dropped to 10% cpu, but the temps stayed in the 85c range, still a bit warm. Folding must use aspects of the cpu (like integer/fp?) that normal workloads don't lean on as much. About the only thing I could do is water cool it, but that'd be a major job since its in a case you can't remove the back of, so I'd have to pull the motherboard.

My 6700K with a huge water cooler on it is keeping up <80C and my 6700 rig with a large air cooler stays around 78C, but I sure would like to get the 3700x into the mix.

Also need more work units! I had all 3 of them on all day the last few days and only caught about 8 hours of work units.

Certainly seems to put a different stress on the CPU than does some of the benchmarks. I can run my 3950x at 4.4 all core and run Cinebench and cpu-z stress tests without issue but doing so with Folding running a job will crash the system.
And agreed on the work units. That 10x supercomputer power spends most of the day twiddling its thumbs. Fully closing and restarting the client seems to have a greater success rate in getting WU for either the CPU and/or GPU vs. just waiting.

 

[shorttack]

I tried it today but saw now reference to covid 19 on the settings, only cancer and alzeimer.
Why is that?

The client program has not been updated to show Covid as a choice. Make sure to set up CPU as a folding client and you are sure to get Covid work. GPU is more mixed.

 

[shorttack]

The CPU client doing mostly Covid work is using the AVX instructions, which are huge power consumers (e.g., 512-bit multiply). I would just let it run. 85C is hot, but the chip has excellent throttling circuitry, and you have not hit throttling yet.

 

[cryoburner]

A quick check of the stats today at Folding@Home reveals the network is now theoretically capable of 1.5 exaFLOPS from CPUs, powered by 4,630,510 CPU cores, and is augmented by 435,563 GPUs.

The network might "theoretically" have that level of computing power, but from the sound of it, most of these computers are sitting idle most of the time with the back-end unable to provide work for them to do, so the actual level of computing power being utilized is far lower. Comparing the network's theoretical performance versus other supercomputers performance doesn't serve much purpose if you don't know how much work these systems are actually doing.

And of course, there's the question of how much of the work that's getting done is even related to coronavirus research, the headlining thing that many people are running the software for. It's probably a rather small portion, compared to other research efforts running on the network.

 

[Unolocogringo]

Cooling tips !!!!
Most stock CPU coolers will make the CPU run very hot when folding. In cases with poor airflow they can overheat very quickly.
Install the biggest Air/Water cooler you can get in your case.

Install the highest CFM fans you can install . You want to get cool air in and hot air out as fast as possible.
I use Aerocool shark fans for high volume semi low noise cooling. 82.6 CFM @26.5db , 1.23mm static pressure. for a 120mm model.
The new black Noctura Industrial line is also a good choice.
Manage air flow. Which will vary by case layout.
If all else fails remove both side panels and point a house fan blowing in the motherboard side.

 

[cfbcfb]

Certainly seems to put a different stress on the CPU than does some of the benchmarks. I can run my 3950x at 4.4 all core and run Cinebench and cpu-z stress tests without issue but doing so with Folding running a job will crash the system.
And agreed on the work units. That 10x supercomputer power spends most of the day twiddling its thumbs. Fully closing and restarting the client seems to have a greater success rate in getting WU for either the CPU and/or GPU vs. just waiting.

Yeah, II figured out the trick of restarting the client, but while I got a solid bunch of WU's a few days ago, they've dried up to the point I'm not running the client anymore. I found many workloads specify 16 core or more, cpu only, or were offered with very short expiration windows and extremely low points, although I could care less about points. I just want to help.

So it seems the troubles are too many clients and not enough WU's, the WU's authors wanting to get their results in 15 minutes vs an hour so as to be overly picky about the iron it runs on and server end overloads issuing and accepting results.

I believe the f@h client uses more of the integer and fpu math execution than your average cpu tester, that tends to be more broad spectrum.

I've also learned a lot about the 3000 series ryzens and cooling. After finding so many people with cooling issues, even with fairly large air cooling setups, and only the folks with higher end water cooling being able to fold without overheating, seems AMD's "chiplet" design changes cpu thermals. Used to be most of the heat came from the center portion of the die, while the ryzen architecture spreads that around the package.

I did a whole day of testing with different coolers. I found the ones with a larger heat sink and a square contact point larger than the die offered a good 10c improvement. Even a very tall Aidos radiator model with a huge fan and a round contact point that didn't cover the edges of the die wasn't any better than the crap cooler that HP put on the cpu. I made sure case temps weren't an issue with the side off and a big fan blowing into the system.

I then read a lot about how PBO works (precision boost, default turned on) and that it may seriously push thermals by over volting and/or overboosting to show better short term performance, but sustained use of PBO can create heat problems. As in running a 3 hour cpu fold vs a 2 minute benchmark. I found a windows power plan called '1usmus universal', made specifically to address cpu thermals while not hitting performance by much. THAT was a bigger deal, giving me a 10C drop in temps while folding.

So with a new much bigger heat sink (1.5" thick instead of .5" thick), a faster cpu fan, and that power plan installed, I went from 92-94C cpu temps down into the low 80C range. I'm okay with 80. Seems AMD is okay with 90+ but I'm not!

 

[scodd]

For almost a year, I thought my 9700k overclocked at 5.0 on all cores all the time was stable. It passed all the tests for full days with flying coolers. Then a month ago I started folding and bam. Kinda funny, looks like I am not the only one. I reset everything to stock for a few days but cranked up the fans. I few days later I started using the Intel tuning util and have had it going at 4.7 all day full since without a single error in folding logs. Just about to hit 2 million for Tom.

 

[scodd]

Also noted, Handbrake gets it hotter than folding for me.

 

[scodd]

The network might "theoretically" have that level of computing power, but from the sound of it, most of these computers are sitting idle most of the time with the back-end unable to provide work for them to do, so the actual level of computing power being utilized is far lower. Comparing the network's theoretical performance versus other supercomputers performance doesn't serve much purpose if you don't know how much work these systems are actually doing.

And of course, there's the question of how much of the work that's getting done is even related to coronavirus research, the headlining thing that many people are running the software for. It's probably a rather small portion, compared to other research efforts running on the network.

Not sure about the rest of the folks folding but for over three weeks I have had a 3930k and 9700k running full and getting very little rest between jobs. I think there was distribution issue going for a bit but that seems to resolved and jobs are being farmed at a very high rate as far as I can tell.

Is anyone slow on getting assignments? Every job I have received has been related to Covid research except right at the beginning a few small jobs related to Arctic micro-biology or something similar came in. Since then the jobs have gotten bigger sometimes take 5 hours or more and are all Covid research.

 

[PBme]

Not sure about the rest of the folks folding but for over three weeks I have had a 3930k and 9700k running full and getting very little rest between jobs. I think there was distribution issue going for a bit but that seems to resolved and jobs are being farmed at a very high rate as far as I can tell.

Is anyone slow on getting assignments? Every job I have received has been related to Covid research except right at the beginning a few small jobs related to Arctic micro-biology or something similar came in. Since then the jobs have gotten bigger sometimes take 5 hours or more and are all Covid research.

I've been getting more jobs in the past week or two but it still can go quiet for a day or so until I restart it. No job is taking that long though. CPU jobs go very quickly (20 mins?) on my 3950x but GPU jobs on my rtx 2070 (non super) seem to take about 4-5x as long. Though they give about <10k points for CPU and 100-200k points for GPU job. No idea how they determine that sort of thing.

 

[scodd]

Jeez, on my 9700k a 4 hour job gets about 35000 points. I should have waited 9 months on my last upgrade and picked up a 3900 or so. Ooh well.

 

[PBme]

Jeez, on my 9700k a 4 hour job gets about 35000 points. I should have waited 9 months on my last upgrade and picked up a 3900 or so. Ooh well.

That isn't too bad. I've not seen any CPU jobs anywhere near than many points. It would take my current CPU jobs about ~2.5 hours to get 35,000 pts with the current jobs that are around 8k. 4 hours isn't bad at all for having 1/2 the cores. But no cpu compares to GPU jobs point wise.

 

[cfbcfb]

I've been getting more jobs in the past week or two but it still can go quiet for a day or so until I restart it. No job is taking that long though. CPU jobs go very quickly (20 mins?) on my 3950x but GPU jobs on my rtx 2070 (non super) seem to take about 4-5x as long. Though they give about <10k points for CPU and 100-200k points for GPU job. No idea how they determine that sort of thing.

The GPU with a proper workload can do far more calculations than a cpu, so those long gpu workloads might take 5-10x longer as a cpu load, hence the higher points.

There are big cpu jobs that can't be done (or the researcher didn't build the work unit to do it) on a gpu. Some of those carry a heavy point load.

However after days of two loaded machines running non stop to get 2 days of nothing and now a few days of a load here and there, I'm dropping out. Seems there's plenty of compute power available. No need for me to add $100 to my electric bill to process some occasional load about myocardial infarctions effects on cheese eating mice.