I'm adding a bit of firepower today. We are closing in on old friends Tekzilla and DSL Reports Team Helix (note the absolute number of work units crunched for those two teams and also the Power Cows compared to the johnny-come-latelys). Hopefully, we'll head back into the top 50 teams.
Re: [TER] Folding@home. Home thread.
- Thread starter Hinton
- Start date
I'm likely to be out in terms of F@H production for the indefinite future, not that I've been contributing much lately (office power limits, I can't run more than about 1200W of load without browning things out when the AC starts).
A friend is doing some increasing work with 3D modeling and could use a substantial workstation upgrade from his ancient Sandy Bridge small form factor PC, so my office heater (i7-3820, GTX980, GTX1060) is getting cleaned off and delivered his way. This means I won't have GPUs for F@H (the 1060 is going his way too as an upgrade for his house PC, as GPUs have gotten very, very stupid lately).
I'm going to replace that capacity with a few more Broadwell class systems that will be working on CPU tasks for other projects. I'm probably going to decommission the Xeon heater too, it's 12C/24T, but not very fast at anything (barely faster than the 4C/8T Broadwell box in Geekbench), and I can get more work done for the power if I move that power budget to something slightly more modern and efficient but still used. For no reason beyond "I think it's neat," I'll probably build a few more of the eDRAM boxes around the i7-5775Cs. I've no real way to tell if that helps or not (I can't find anything that monitors the eDRAM use and I'm tired of digging at performance counters), but it certainly won't hurt! There are some Xeons with eDRAM but they're impossible to find.
So, still will be doing compute, but my GPUs are going to get better used as GPUs by someone actually doing pixel pushing with them. Just need to figure out how to run Electric Sheep on a Pi for ambient video...
A friend is doing some increasing work with 3D modeling and could use a substantial workstation upgrade from his ancient Sandy Bridge small form factor PC, so my office heater (i7-3820, GTX980, GTX1060) is getting cleaned off and delivered his way. This means I won't have GPUs for F@H (the 1060 is going his way too as an upgrade for his house PC, as GPUs have gotten very, very stupid lately).
I'm going to replace that capacity with a few more Broadwell class systems that will be working on CPU tasks for other projects. I'm probably going to decommission the Xeon heater too, it's 12C/24T, but not very fast at anything (barely faster than the 4C/8T Broadwell box in Geekbench), and I can get more work done for the power if I move that power budget to something slightly more modern and efficient but still used. For no reason beyond "I think it's neat," I'll probably build a few more of the eDRAM boxes around the i7-5775Cs. I've no real way to tell if that helps or not (I can't find anything that monitors the eDRAM use and I'm tired of digging at performance counters), but it certainly won't hurt! There are some Xeons with eDRAM but they're impossible to find.
So, still will be doing compute, but my GPUs are going to get better used as GPUs by someone actually doing pixel pushing with them. Just need to figure out how to run Electric Sheep on a Pi for ambient video...
Welp, with the shortage of work units, I've installed BOINC (first time ever) and semi-randomly chose Einstein for the project based on the AMD support as discussed in the folding forum thread about the lack of work units. I can do a work unit about every two and a half minutes at about 3400 points per work unit. Does math (takes off socks and shoes). That's about 1,800K ppd on the fast box. Slow box gets about 640K ppd.
I haven't been able to run my GPU as often in the high heat lately, but I have been able to keep the CPU fed under Windows at least. A Ryzen 3700X under an 8-thread load is relatively easy to manage in the office, but ramping up the 1070 Ti means I'm fighting to help keep the office from getting too toasty. If it cools down enough in the evenings, though, I'll let the GPU chew through a few work units before turning it off in the morning.
@Burned - thank you. I never expected that. When I started SuicideBiker was the only team memeber above 1B points, and that seemed out of reach. Now we have six members over that milestone 
Gave away my 1060 to my son a few months ago, because he literally couldn't buy a decent GPU for anything approaching a sane price, which dropped my output by about 600K ppd. At least it helped the cooling situation.
Was messing with my home network a few weeks ago, and discovered the F@H client has to have internet connectivity to recognize my 2060, and likely other GPU's as well. You obviously have to have internet to do any work, but the error message was along the lines of "GPU Disabled" which was a little alarming at first.
@For_Caitlin - congrats on the jump in output.
Gave away my 1060 to my son a few months ago, because he literally couldn't buy a decent GPU for anything approaching a sane price, which dropped my output by about 600K ppd. At least it helped the cooling situation.
Was messing with my home network a few weeks ago, and discovered the F@H client has to have internet connectivity to recognize my 2060, and likely other GPU's as well. You obviously have to have internet to do any work, but the error message was along the lines of "GPU Disabled" which was a little alarming at first.
@For_Caitlin - congrats on the jump in output.
@hutchm That's me. I was folding on an old MacBook Pro and not really paying any attention to DC. I had to send it back to the employer it came from. I had built a new desktop that I got a video card for and started folding on it. Then I built a way overkill game rig for my new hobby (Zwift) and won the Newegg lottery to get one of the new AMD cards. Took a recent hiatus to Einstein@Home, but am back full time to folding.
Team Egg Roll had a good week last week. We passed Brookhaven National Lab and old rivals DSL Reports Team Helix and Tekzilla to climb in to 56th place with almost 38 billion points racked up. We're tracking pretty close to project overall trends. The big covid ramp up in users is slowly falling by the wayside. We're in 52nd place in daily production, so there is still room for us to move up in the ranks over the next two or three years. We're in 26th place in work units completed, a testament to our longevity if nothing else. For comparison, in BOINC we're 48th overall and drifting downwards, and our friends at Team Prime Rib over in the GIMPS project are in 26th place. I don't think there are any other notable active DC platforms.
Sadly a confluence for factors is taking me offline. I returned to folding because of COVID and never expected to contribute as much as I have, but looks like I will be taking an indefinite pause.
It's been fun, happy folding y'all.
It's been fun, happy folding y'all.
The weather has been a lot cooler here lately, so I've been able to get some more overnight folding done on my trusty 1070 Ti (as id4xdmrh, the old randomly generated user name I got a LOOOOOONG time ago). It's good for about 1.3-1.5 million points per day (run for about 1/3 of a day these days) since the switch to CUDA, and once prices for more modern GPUs get back down to somewhat sane levels I'll likely be upgrading it to something at least as fast as a RX 6700 XT or 3060 Ti/3070.
Once fall and winter hit I will run it more often since the heat is a GOOD thing that time of year in The Frozen North, and actually keeps my office quite comfortable.
Once fall and winter hit I will run it more often since the heat is a GOOD thing that time of year in The Frozen North, and actually keeps my office quite comfortable.
So I have this M1 Mac Mini in the living room sitting idle 23 hours a day. If I could find my old F@H account information and figure out how to install the client I think I'd give it a shot. Any pointers?
Find a project that has M1 native binaries, and... I think there are some issues with swapping on the Rosetta tasks if you let all the cores run, otherwise it's a great little compute box. I don't recall if that was an issue with native tasks or x86 translated, but no point in using it to run x86 jobs when you can run ARM native jobs. Pretty sure F@H doesn't have a M1 client yet.
Well, it's finally getting chilly enough in the mornings that I can run a work unit or two before my office gets too warm. Keep an eye out for my ~250k/day points and all
Well, I've been asleep at the wheel and didn't notice my ppd falling off a cliff. Turns out I needed to update my NVidia drivers just like the folding forum sticky thread says.
Thanks Jimbo. It will be interesting to see what the future holds, not only for Folding, but for DC in general. Total project PPD looks to be down about 50% from its covid peak, but still well above its baseline.
I seem to be getting slightly improved performance with the 5xx Nvidia drivers over the 4xx drivers. Is anyone else noticing this?
I got 26M PPM vs 24M PPM.
I got 26M PPM vs 24M PPM.
I'm going to install the V5 drivers here in just a minute. I've been disappointed with the AMD drivers. I play one game on that machine and its been unstable (avatar strobing and random hangs and crashes) so I am installing the latest drivers as soon as they are available. My PPD production has been reduced about 10% or so.
As a team, we've been holding a steady position at #56. Our output has ramped up. We have several new users and several others who look like they got nice Christmas presents last year. We've moved up nicely to 42nd in daily team production. Our old friends the Dutch Power Cows are in our sights, and we should pass them in the next few months, but we have other teams hot on our tail. It looks like we may slowly claw our way up a few spots over the next year or so. The overall health of the project appears good too and has been holding steady for the last year.
As a team, we've been holding a steady position at #56. Our output has ramped up. We have several new users and several others who look like they got nice Christmas presents last year. We've moved up nicely to 42nd in daily team production. Our old friends the Dutch Power Cows are in our sights, and we should pass them in the next few months, but we have other teams hot on our tail. It looks like we may slowly claw our way up a few spots over the next year or so. The overall health of the project appears good too and has been holding steady for the last year.
The 3060 Ti I grabbed last September has more than doubled my Folding@Home productivity for only 50 W more power draw compared to the 1070 Ti it replaced, and combining that with a winter that could best be described as "Soviet-ass unrelenting cold" meant that I was able to fold pretty much constantly from December to March.
I know that I have gained points much more quickly than ever before, and during the summer I'll still likely be CPU folding for a good chunk of the day since my 3700X is still pretty efficient even in warm weather. Its performance is MUCH better under Linux than Windows 10, but I'm still a bit annoyed that no one has managed to port the Linux version of FAHControl to Python 3 yet.
I know that I have gained points much more quickly than ever before, and during the summer I'll still likely be CPU folding for a good chunk of the day since my 3700X is still pretty efficient even in warm weather. Its performance is MUCH better under Linux than Windows 10, but I'm still a bit annoyed that no one has managed to port the Linux version of FAHControl to Python 3 yet.
Just a quick update... we've fallen one rank to #57, as Missouri S&T passed us. The good news is at our current course and speed we should continue to climb slowly in the ranks over the next year and pass old rivals the Dutch Power Cows. Currently, we're about 40th in weekly production, so over the long term we should continue to climb. Not bad for a website team that gets no pimpage from its owners. Personally, I topped the two billion point mark and entered the top 1000 of the project. As you can probably tell, discussion around here (and really everywhere) on distributed computing is pretty moribund. I suppose it will remain so unless something happens to really capture the public's imagination, like SETI once did and to a lesser extent Covid. Fold on fellow egg-rollers.
I have no Folding rigs running right now. My CPU time is going to ClimatePrediction.net, though they keep running the work unit pool dry too. I hear rumors of new WUs out of them, though! Currently, my Linux boxes are pretending to be old Mojave Macs with VMs to run the 32-bit Intel/MacOS WUs that were available.
I keep wanting to add another F@H rig, but I don't want to drop the coin on GPUs right now.
I keep wanting to add another F@H rig, but I don't want to drop the coin on GPUs right now.
I had a dream, a dream of breaking into the Team Egg Roll top ten that has now been crushed by a bunch of upstarts racing up the rankings (no, it has nothing to do with only running part time on a GTX 1070 or selling off some other hardware last year...).
I had to back-level my AMD driver to be able to Zwift. I think it actually helped my production. I'm pretty disappointed in the whole AMD graphics experience (and in Zwift QA and support, but that's another story). Anyway, I noticed the bottom fell out of cryptocurrency, so maybe that will help with hardware availability.
To expand on my previous remarks from May As some of you know, Ethereum can't be economically mined anymore (there was a front page article about it the other day). That is a lot of graphics card power out there looking for something to do. Will some of it start folding? The daily points trend has started back upwards again. Theoretically, the bottom should drop out of the used card market. One might be able to build a pretty significant pharm for cheap.
As some of you know, Ethereum can't be economically mined anymore (there was a front page article about it the other day). That is a lot of graphics card power out there looking for something to do. Will some of it start folding? The daily points trend has started back upwards again. Theoretically, the bottom should drop out of the used card market. One might be able to build a pretty significant pharm for cheap.I should poke around eBay. I just got a new house PC online that I don't need, but it could fit a GPU or two. It's got a 1080Ti in it right now.
F@H on Linux seems pretty painfully out of date, though. The FAHControl app won't install without a lot of hacking on newer Ubuntu distros...
F@H on Linux seems pretty painfully out of date, though. The FAHControl app won't install without a lot of hacking on newer Ubuntu distros...
Recent instability with my Raedon box have frustrated me to the point that I've moved it to Einstein while I try to figure out what combination of drivers, client code and/or work units is behind this. The crashes have gone from occasional to near constant.
First some good news, the RTX 4090 seems to get up to 22 million Points Per Day.
sadly the Intel cards cannot currently fold.
"To explain what that means in practice, folding mostly uses single precision floating point calculations. Those are also called 32-bit or FP32 calculations. A small amount of the calculations need more precision than can be obtained with 32-bits, so double precision 64-bit (FP64) calculations are used where needed.
How this applies to the Intel graphics is that when first supported the iGPUs from their 9th generation supported FP64 in hardware. For whatever reason there was not a designated 10th generation, the 11th generation iGPUs also supported FP64. With the 12th generation GPU architecture Intel removed the FP64 hardware support and added code in the driver to emulate FP64 calculations with software instructions. From early information out on the just released Arc GPUs they are using the same approach to support FP64. Tests done using folding gave inconsistent results, or caused the folding core to crash."
sadly the Intel cards cannot currently fold.
"To explain what that means in practice, folding mostly uses single precision floating point calculations. Those are also called 32-bit or FP32 calculations. A small amount of the calculations need more precision than can be obtained with 32-bits, so double precision 64-bit (FP64) calculations are used where needed.
How this applies to the Intel graphics is that when first supported the iGPUs from their 9th generation supported FP64 in hardware. For whatever reason there was not a designated 10th generation, the 11th generation iGPUs also supported FP64. With the 12th generation GPU architecture Intel removed the FP64 hardware support and added code in the driver to emulate FP64 calculations with software instructions. From early information out on the just released Arc GPUs they are using the same approach to support FP64. Tests done using folding gave inconsistent results, or caused the folding core to crash."
Jimbo, you have your finger on the pulse of the folding forum... any ideas where I should start looking for my Radeon RX 6900XT stability issues? It of course chugs along just fine on Einstein, so its probably not hardware.
I don't run an AMD GPU,my last was a ATI HD 2600 pro from 2007. But newer drivers have newer bugs, so try new drivers from AMD.
Yeah, I've fooled with different AMD drivers to no avail. And just now, my NVIDIA box is getting an error downloading the new core. So its on Einstein for the time being also.
Sadly, F@H seems to have no test system, some researcher was editing the live system, badly.
We're into the winter now here in Winnipeg, which means that I'm helping offset my home's natural gas greenhouse gas emissions with some hydroelectric power from my 3060 Ti. It's very good at keeping the office toasty warm, and cranks out work units very quickly. At this rate I may crack the billion point mark by the end of the winter!
Not sure F@H makes much sense anymore. AlphaFold dominates protein folding. Old approaches are completely obsolete. Seems like a waste of electricity. And if there are empty spaces where AlphaFold doesn't answer the question, it's just a matter of time.
Firstly, by way of context, I've been contributing to Ars DC campaigns since 1993 (SETI@home back then), and folding on my iMacPro for the past few years. I honestly don't have the time or patience to develop the expertise necessary to take advantage of the GPU in this system (Radeon Pro Vega 56), which is a shame because it was a pretty impressive GPU when it was new.
But
So I'm looking for a good biomedical DC problem to point my spare CPU cycles at; ideally something with a good macOS client that can use the GPU.
My office is cold and the university doesn't track power consumption at all
But
I'm definitely inclined to agree with this.
So I'm looking for a good biomedical DC problem to point my spare CPU cycles at; ideally something with a good macOS client that can use the GPU.
My office is cold and the university doesn't track power consumption at all
SETI's first release was in 1999.
As the founder of TER, I'm a little sad to say I think it's time to hang it up as a project.
I just drove through Wikipedia's list of active DC projects:
en.wikipedia.org
TBH, not many of them seem very interesting. My criteria for "interesting" is "life science" + "not a waste of time" + "trying to accomplish something useful".
Rosetta@home has the same problem as F@H, I think -- it's not worth doing anymore due to AlphaFold. Trying to improve on outmoded methods of predicting protein folds and misfolds seems like it's trying to optimize the horse-and-buggy in the age of the automobile.
As the founder of TER, I'm a little sad to say I think it's time to hang it up as a project.
I just drove through Wikipedia's list of active DC projects:
List of volunteer computing projects - Wikipedia
TBH, not many of them seem very interesting. My criteria for "interesting" is "life science" + "not a waste of time" + "trying to accomplish something useful".
Rosetta@home has the same problem as F@H, I think -- it's not worth doing anymore due to AlphaFold. Trying to improve on outmoded methods of predicting protein folds and misfolds seems like it's trying to optimize the horse-and-buggy in the age of the automobile.
Last edited:
The two methods solve different problems. Dr. Bowman addresses this on the website. Alpha predicts a dominant end-state. F@H explains how you get there. If you're looking for a cryptic pocket on a SARS-CoV spike protein, alpha's not going to help much. F@H crunching is still producing results published in peer reviewed scientific journals, so reports of its demise might be greatly exaggerated.
I must have missed when it happened exactly, but I just passed over a billion points in Folding@Home! I'm in the top 0.1% of all donors according to the statistics, and it's been a very long time folding to get to this point. Once I got started with GPU folding on my faster cards since 2019 it has helped immensely, but I've been patiently chugging away on a variety of CPU and GPU clients since at least 2006 or 2007. I'm going to keep going, and I am happy to have contributed to so much scientific testing over the years!
I'm probably going to end my sojourn in Einstein pretty soon and rejoin the fold. We made pretty good headway for awhile, but the top 40 is going to elude us I think.