Folding Community Board

[Shlouski]

7 hours ago, RollinLower said:

have you tried cleaning it? alcohol so far doesn't seem to help much. it's the worst kind of sticky goo too

 

O yeah it's a pain, had many cards do this to me.

 

As soon as you wipe off a little on to whatever you are using to clean it you end up just spreading it around. On metals I have used paint thinner to remove this stuff, also works well on heatsinks to remove that stain thermal paste can leave behind. I can't recommend paint thinner and you use at your own risk, some can leave residue behind so always clean again with alcohol after using it, I would be very careful using it on paints and never on plastics.

 

The safe option is dabbing and short wipes using tissue paper folding it over and dabbing again with a clean spot, don't use alcohol as it makes it harder for the oil to stick to the paper. After you have removed the bulk of it use a lint free cloth and alcohol to finish the job.

[Favebook]

15 hours ago, GOTSpectrum said:

this is the badge request thread, go to here for community support for the folding team

 

 

@GOTSpectrumThis is the thread that you linked, not the badge request thread

[GOTSpectrum]

1 hour ago, Favebook said:

@GOTSpectrumThis is the thread that you linked, not the badge request thread

Listen... I'm not gonna lie, a lot of morphine was taken last night 

[TVwazhere]

3 hours ago, Favebook said:

@GOTSpectrumThis is the thread that you linked, not the badge request thread

2 hours ago, GOTSpectrum said:

Listen... I'm not gonna lie, a lot of morphine was taken last night 

Nah that's on me. I moved the comments out of the Request thread 

[GOTSpectrum]

32 minutes ago, TVwazhere said:

Nah that's on me. I moved the comments out of the Request thread 

ohh... so it wasn't the morphine... 

 

we will let it off... for now 

[GOTSpectrum]

BTW guys, I'm hoping to work on the first post of this thread in the near future, so if you have any ideas what you would like to see please do let me know 

[Gorgon]

19 hours ago, Gorgon said:

Freezing rain here and the power went out six hours ago. No idea when they’re going to restore it.  Cellular data is unusable but I remembered I just took a 1500VA UPS out of service the other day so I plugged the router and modem into it and voila!  Connectivity

25 hours later and the power is back. A least it's not too cold here overnight and during the day so the house only got down to 16.5C. Heats back on and I started up the folding rigs early to help warm things up.

[GOTSpectrum]

I have reached the 5b point threshold, this is what we like to see 

 

[Bitter]

2 hours ago, Gorgon said:

25 hours later and the power is back. A least it's not too cold here overnight and during the day so the house only got down to 16.5C. Heats back on and I started up the folding rigs early to help warm things up.

I had a breaker pop and a couple GFI outlets across the house trip during some bad weather here, I think there must have been some really bad low voltage, my folding stuff went offline and half the house network with it. I haven't gotten the folding PC back up yet since I need to add the 3rd card, a project for this Saturday perhaps.

[GOTSpectrum]

 

[Mxyzptlk]

22 hours ago, Shlouski said:

 

I'm running at 50% PL on all of the 

Yeah I noticed the power consumption didn't chance much when increasing the PL, but setting 50 PL doesn't give me problems on single gpu systems, also increasing the PL on that 5 gpu rig to 100% does little to nothing for me. 

 

 

I can keep decently close to normal PPD all the way done to 60ish %.  Once I go below that my PPD Tanks. 
 

Have you checked your 3.0x4 slot performance compared to 3.0x8 or higher?  You have some great hardware, would be great to see it fly! 

[Gorgon]

23 hours ago, Shlouski said:

... I'm running at 50% PL on all of the cards, this is because even at 100% PL the cards usually draw 150-170 watts on most WU's which is very close to the 50% PL anyway, and I have observed Alheimers WU's ppd decrease by less than 20% of a 100% PL. All other setting are at default.

 

A few times I have tried increasing the power limit but this has little to no effect, likely due to most WU's only pulling around 50% PL, 150-170 watts. ,,,

 

46 minutes ago, Mxyzptlk said:

I can keep decently close to normal PPD all the way done to 60ish %.  Once I go below that my PPD Tanks. 
 

Have you checked your 3.0x4 slot performance compared to 3.0x8 or higher?  You have some great hardware, would be great to see it fly! 

Yeah - I'd try upping the PL a bit. I've found on Turing and Ampere that I have better stability of yield using a higher PL and locking the maximum graphics clock but I have no idea what the sweet spot is for Ada.

 

I think @Mxyzptlk or you found somewhere around 2.1-2.2GHz so:

nvidia-smi -i 0 -pl 200
nvidia-smi -i 0 -lgc 0,2200

or something similar.

[Shlouski]

1 hour ago, Mxyzptlk said:

Have you checked your 3.0x4 slot performance compared to 3.0x8 or higher?  You have some great hardware, would be great to see it fly! 

 

I have done little testing with slot configurations so I don't know when 3.0 x4 would start to cause a bottleneck, I but I have seen expected performance on this rig when it still only had 2 or 3 cards in it also running at 3.0 x4.

 

Interesting side note, I did find out that a 6600xt is around the point a 2.0 x1 slot starts to bottleneck, on a 2.0 x1 slot it was able to do a solid 1.2 mil ppd at very close to 100% usage.

Radeon RX 6600/6600 XT/6600M Folding@Home PPD Averages, Power Consumption & Research Projects (lar.systems)

 

I'm sure it's not this simple, but if a 0.500 GB/s connection can support cards capable of 1.2 mil ppd, perhaps a 1.000 GB/s 2.0 x2 or 3.0 x1 could support cards capable of 2.4 mil ppd, so perhaps the limit of 3.0 x4 is 9.4 mil ppd. My experience tells that this is not the case and that it doesn't scale like this, I think for every doubling of bandwidth you get more than double the performance limit. 

 

I will run some tests on my 4090 now and see if I can come up with some rough info on the limitations of various bandwidths.

 

 

[Shlouski]

On 4/7/2023 at 5:24 AM, Shlouski said:

I will run some tests on my 4090 now and see if I can come up with some rough info on the limitations of various bandwidths.

 

Well I did some testing.

 

Unfortunately I'm only getting cancer WU 18717 to test with at the moment. I can see that even pcie x4 2.0 has enough bandwidth to hit the average ppd score for this WU, but this might not be the case for larger WU's.

 

This also leads me to believe that bandwidth is likely not the cause of the issues I'm having with my folding rig. 

[LAR_Systems]

Hi everyone, have a little bit of an Easter update for https://folding.lar.systems/ that will hopefully help with some inter team competition and possibly that of running events in the future.

 

I have been working on a Folding League stats system that can be used year round and/or during folding events to provide as real time as possible leader board based on a number of different factors which I will explain below.

 

ROLLING 24 HOUR PPD PRODUCTION RANKS
FOR ACTIVE TEAM MEMEBERS

https://folding.lar.systems/league/team?id=223518

 

This displays folders on the team 24 hour production PPD and change as it relates to a previous day including a "Day League" which is a named grouping based on your global to all folders world wide PPD production percentile.  These values update throughout the day as folding file updates are available to be processed which are checked every hour with the day ending / starting at midnight eastern time.

 

 

ROLLING MONTH TO DATE POINTS TOTAL
FOR RANKED FOLDERS - APRIL 2023

https://folding.lar.systems/league/team_ranks?id=223518

 

This displays the teams users rolling overall monthly total and rank inside the team overall much like the way "folding month" stats have been typically done manually.  The stats update throughout the day as F@H provides new stats files.   In addition to the overall rankings each folder on the team will have a days active rank and a "League" they belong to that is based on their daily production percentile group, the number of days active in the month to date they have folded averaged out.   For example if you're in the All Star group and fold every day of the month, you will stay in the All Star group and ranks... if you miss some days, have a machine go down... that will drop your percentile of production average and you might slide in to a different league which are ranked highest to lowest left to right on the NAV.

 

 

ROLLING MONTH TO DATE POINTS TOTAL
FOR ALLSTAR LEAGUE FOLDERS - APRIL 2023

https://folding.lar.systems/league/team_leagues?id=223518&league=allstar

 

You can then look at individual Leagues tabs to see the monthly user ranks inside the team, days active PPD scores etc.  These groups of percentile based folding users allows those that can only fold part time, only have certain tiers of equipment to fold on etc. be grouped in with similar users of similar production still have competition which could be used for prizing groups during events etc.

 

 

https://folding.lar.systems/league/team_leagues?id=223518&league=majors

 

https://folding.lar.systems/league/team_leagues?id=223518&league=aaa

 

ETC... ETC... 

https://folding.lar.systems/league/team_leagues?id=223518&league=aa

https://folding.lar.systems/league/team_leagues?id=223518&league=minors

https://folding.lar.systems/league/team_leagues?id=223518&league=rookie

https://folding.lar.systems/league/team_leagues?id=223518&league=peewee

 

Last if you click on a folders name there is now a profile which will show that folders info for the overall lifetime, month, league and daily production values that were used to calculate the above divisions.

 

https://folding.lar.systems/league/user?name=Mxyzptlk&team=223518

 

 

You can find your user name in the list of active folders in the overall 24 hour or league production lists and click to find it.

 

Enjoy and have a great weekend.

 

 

 

 

 

 

 

 

 

[Mxyzptlk]

@LAR_Systems Awesome work! Just bought you a few coffee's.

[LAR_Systems]

9 hours ago, Mxyzptlk said:

@LAR_Systems Awesome work! Just bought you a few coffee's.

Thanks, I really appreciate it!
This thing runs on caffeine big time.  

[dpt17]

On 4/7/2023 at 10:46 AM, LAR_Systems said:

Hi everyone, have a little bit of an Easter update for https://folding.lar.systems/ that will hopefully help with some inter team competition and possibly that of running events in the future.

...

 

Nice.  I feel like I've reached a point where I'll have to make a change to keep pace.  I'm curious to see how well the 4070 does in F@H.  Assuming they are buyable at $600 retail, grabbing two of those might exceed a 4080 performance and approach 4090 performance with Wattage 2 x 186 = 372W ... more than the 4080 but less than the 4090.  My folding box is i9-10850k, Asus z590, 3080 Ti + Intel 10Gbe nic (so x16/x0/x4), 850W psu.  My choices are

1. Grab lower power 4090 (that will work with 850W psu) and probably leave it running in PCIe 3.0 x16 mode: ~$1600
2. Grab two smaller width 4070s and an i7-11700k (to open up PCIe 4.0 x8/x8): ~$1400
3. Grab something that's good for folding and isn't too gimped by PCIe 3.0 x4: ?

Is there a good card to run in PCIe 3.0 x4 mode?  I have a spare Razer Core X eGPU and my mobo has Thun4 support.

[Gorgon]

1 hour ago, dpt17 said:

... My choices are

1. Grab lower power 4090 (that will work with 850W psu) and probably leave it running in PCIe 3.0 x16 mode: ~$1600
2. Grab two smaller width 4070s and an i7-11700k (to open up PCIe 4.0 x8/x8): ~$1400
3. Grab something that's good for folding and isn't too gimped by PCIe 3.0 x4: ?

Is there a good card to run in PCIe 3.0 x4 mode?  I have a spare Razer Core X eGPU and my mobo has Thun4 support.

The sad fact is that at current prices the 4090 is still the best "value" if you look at Shaders/$ it even slightly beats out 4070s assuming the 40470 is available at the MSRP. The lineup in terms of best value at current (Canadian) Street Prices is: 4090, 4070, 4070ti, 4080.

 

If you look at (Shaders x BaseClock)/$ the lineup shifts slightly to: 4090, 4070ti, 4070, 4080 due to the 4070s (rumored) gimped Base Clock. But this is less of a valid metric as you should, if you care at all about efficiency, be running your GPU at a reduced Power-Limit and/or restricted Maximum Clock.

 

But, as you say, being able to run 2 x 2-slot 4070 GPUs in a Rig offers a compelling setup which would be less prone to having their Yield being completely gimped by small atom-sized or under-performing WUs than single 4090.

 

I suspect PCIe3 x4 should be sufficient for most cards, even a 4090 but I haven't been able to test this but @Shlouski or @Mxyzptlk might be able to better answer that.

[Shlouski]

47 minutes ago, Gorgon said:

The sad fact is that at current prices the 4090 is still the best "value"

 

48 minutes ago, Gorgon said:

If you look at (Shaders x BaseClock)/$ the lineup shifts slightly to: 4090, 4070ti, 4070, 4080

 

49 minutes ago, Gorgon said:

2-slot 4070 GPUs in a Rig offers a compelling setup which would be less prone to having their Yield being completely gimped by small atom-sized or under-performing WUs

 

This is spot on. 

 

Perhaps folding at home could update the client to be able to run multiple WUs in parallel on higher tier gpu's. 

 

I have been trying to solve this problem with gpu paravirtualization, unfortunately I'm a complete noob at virtualization especially when I have to move off windows. I managed to passthrough a 4080 in a hyper-v windows machine, it worked for gaming but Fah wasn't able to detect a cuda or opencl supported card. 

 

@Gorgon

Do you know of any paravirtualization software that might be able work with Fah? 

[dpt17]

19 minutes ago, Shlouski said:

I have been trying to solve this problem with gpu paravirtualization, unfortunately I'm a complete noob at virtualization especially when I have to move off windows. I managed to passthrough a 4080 in a hyper-v windows machine, it worked for gaming but Fah wasn't able to detect a cuda or opencl supported card. 

Thanks for the advice.

 

Virtualization for Nvidia consumer grade GPUs is painful.  From everything I read, Nvidia blocks virtualization of GTX/RTX series cards and that you have to have the quadro / professional grade cards (and drivers!) to do this.

 

In the past, I was working on a project that got me to try a multi-gpu system build.  I was able to get passthrough to work with Linux QEMU virtualization to a Linux guest VM; Ubuntu 20.04 host OS took 1 gpu, Ubuntu 20.04 guest OS took another gpu.  Trying a Win10 guest OS failed with passthrough.  At that time (~15 months ago), I was not able to get passthrough working on Windows.

 

F@H definitely supports multiple, distinct GPUs installed on 1 system it's just that, as your comment implies, you can't carve up a single card to handle multiple work units.  Each gpu gets a fah_core process to handle its job.

[justpoet]

Based on pricing, assuming you can get good 4070 cards at MSRP or below, then 2 of those may be the way to go.  It lets you use one for gaming, split them into different systems later, and not lose from smaller WUs that can't make use of the full 4090 (though there may be less and less of these as time goes on).  You also don't lose both if something goes bad on a card.

 

That being said, the best "overall" value will be the 4090, since you presumably keep it for longer, work units likely keep getting larger for more powerful cards on average, and it still should do more even running efficiently down near the other cards for power output.  You'll be less likely to feel you need/want to replace it as soon as you would the 4070s, so for longevity sake, the 4090 wins.

[Shlouski]

1 hour ago, dpt17 said:

F@H definitely supports multiple, distinct GPUs installed on 1 system

 

This I know. 

 

1 hour ago, dpt17 said:

you can't carve up a single card to handle multiple work units.  Each gpu gets a fah_core process to handle its job

 

I hope this is not the case.

 

I have read that people have be able to get cuda working on both the host and vm at the same time, even using hyper-v as I did. I have the feeling that it might just be a driver issue or  pci root device id conflict, it could be anything really as I'm way out of my depth. 

[GOTSpectrum]

3 minutes ago, Shlouski said:

 

This I know. 

 

 

I hope this is not the case.

 

I have read that people have be able to get cuda working on both the host and vm at the same time, even using hyper-v as I did. I have the feeling that it might just be a driver issue or  pci root device id conflict, it could be anything really as I'm way out of my depth. 

Hmm... FAH uses a GPU whitelist to assign WUs. I'm not sure how that would work with assigning WUs to vGPUs

[Pezui]

It has been quite a while since I have posted. I am still folding with y'all. Currently I have 6th place on the team and 152nd place overall. I am thinking about snagging a new video card, the 3090's are doing "fine" but the numbers the 4090 put up just crush it. I have space to add a 4090 probably though the custom loop will make it clunky to get fitted up proper. Anyways just wanted to stop by and say hello, still bummed I missed folding month back in November and hope to do stuff with the team more engaged again soon.