Folding Community Board

[Gorgon]

21 hours ago, jctappel67 said:

There lies the the rub. I have an SFX power supply in this computer, with only the 2 PCIe outlets. I could make use of the daisy chain connectors, but that's just too janky for my blood. Not even sure if the other would fit without choking out the chonky new GPU anyway (2.7 slots according to MSI).

Doubling up on the power cables would work when power limiting the cards.

 

Not fitting is an issue. You'd likely have to run the 3070ti in a lower slot which is not ideal nor possible on many 7-slot chassis and would cook the. 2070s. I was trying to find 2-slot 3070s but EVGA's run of the XC (Non Ultra) was extremely limited which in retrospect makes sense. If people will buy anything then best to make only the expensive SKUs. The issue is that for JUST power-efficient folding the extra boost isn't needed so there's no point buying something with a higher base clock & beefier cooling solution.

 

21 hours ago, jctappel67 said:

This is why I kept the old case & PSU around anyways, so it's all good.

And it's always nice to build a new system and if your dedicating it to just F@H you can get away with a Pentium or Ryzen 3/5 and a modest amount of RAM to keep the cost down. Or you might be able to find a deal on a used SLI capable (i.e. x8/x8) motherboard & CPU.

[jctappel67]

I know I've been posting a lot lately, but I just had a (good) holy shit moment. I was playing some Far Cry 5 at 1440p Ultra, and I was wondering why my frames were so low, but I capped the framerate at 60 and went on with the session. When I quit playing, I realized that I had left the card folding the whole time I was playing. This thing's an absolute monster, maintained a steady 60 fps, and continued folding at about 2.25 Million PPD, that's insane. My mind is blown. Have a good day.

 

[LAR_Systems]

Team feelings on the LTT skin I might call RGB Lambo?

Team info, stats, badge level shown when you mouse over the logo when live, hidden while not interacting, showing neon LTT logo.

[rkv_2401]

2 hours ago, LAR_Systems said:

Team feelings on the LTT skin I might call RGB Lambo?

Looks amazing!

Maybe more purple around the hardware stuff (the hardware info, PPD average, slot performance etc)? That would really make it pop out!

 

Can't wait for the timer. I know @Gorgon has shared some scripts to automate that, but it would be easier and more accessible to people in the folding community to have it be a part of a web client.

[LAR_Systems]

8 hours ago, rkv_2401 said:

Looks amazing!

Maybe more purple around the hardware stuff (the hardware info, PPD average, slot performance etc)? That would really make it pop out!

 

Can't wait for the timer. I know @Gorgon has shared some scripts to automate that, but it would be easier and more accessible to people in the folding community to have it be a part of a web client.

Thanks for the feedback, made some changes for that Lambo pop.

And yeah, the timer / time of use stuff is a request I have seen in a ton of forums from users old and new, I'm surprised they (FAH) don't have it with an easy mode in the default client cause it has to hurt the number of people willing to fold that don't want to rack up a power bill and heat during peak hours.

 

I know testing it, it's been nice going in to a cool office in the morning with my main system already wrapped up for an hour or two.

 

[Lightwreather]

31 minutes ago, LAR_Systems said:

-snip-

That looks beautiful.

[Gorgon]

I got one of those monster 420,000 Atom 18042 WUs today and while I was waiting for it to finish I played with power limits and got some interesting data.

I'm running HfM.net and it's configured to calculate the performance over the last 3 frames so I adjusted the Power Level of the RTX 2070 Super it was running on from 125 (minimum) to 240W (maximum) and recorded the Credit and Time-per-Frame (TPF) after 4 frames.

So while you can run your GPUs "Balls to the Walls" your rigs will run cooler, quieter and most efficiently at the lowest power limit your card supports.

[rkv_2401]

18 hours ago, Gorgon said:

I got one of those monster 420,000 Atom 18042 WUs today and while I was waiting for it to finish I played with power limits and got some interesting data.

I'm running HfM.net and it's configured to calculate the performance over the last 3 frames so I adjusted the Power Level of the RTX 2070 Super it was running on from 125 (minimum) to 240W (maximum) and recorded the Credit and Time-per-Frame (TPF) after 4 frames.

 

So while you can run your GPUs "Balls to the Walls" your rigs will run cooler, quieter and most efficiently at the lowest power limit your card supports.

There used to be a blog, greenfoldingathome, that tracked all this for various cards, CPUs, and had a table of efficiency (PPD/$ and PPD/watt, from which you can derive a combination of the two) for all components tested. I don't think it's been updated in a while, however. Still read it from time to time to see just how efficient the -60 series cards were for their time.

 

22 hours ago, LAR_Systems said:

Thanks for the feedback, made some changes for that Lambo pop.

This looks amazing.

[GOTSpectrum]

Little update for you all...

 

 

[jctappel67]

On 4/20/2022 at 12:21 PM, Gorgon said:

I got one of those monster 420,000 Atom 18042 WUs today and while I was waiting for it to finish I played with power limits and got some interesting data.

I'm running HfM.net and it's configured to calculate the performance over the last 3 frames so I adjusted the Power Level of the RTX 2070 Super it was running on from 125 (minimum) to 240W (maximum) and recorded the Credit and Time-per-Frame (TPF) after 4 frames.

So while you can run your GPUs "Balls to the Walls" your rigs will run cooler, quieter and most efficiently at the lowest power limit your card supports.

Tested this while running WU 18201. Cut cards power limit to 65% (~200W). It's running about 8 C cooler with a loss of only 300k ppd. I'll take it if the darned thing doesn't make my room into an oven anymore.

[Gorgon]

2 hours ago, jctappel67 said:

Tested this while running WU 18201. Cut cards power limit to 65% (~200W). It's running about 8 C cooler with a loss of only 300k ppd. I'll take it if the darned thing doesn't make my room into an oven anymore.

Yes, GPUs like CPUs these days are designed to run near the top end of the voltage-frequency curve just a “titch” below the point of instability to account for variations in quality. The result is they are horribly inefficient when run at stock settings.

 

it’s not an issue when gaming for a few hours a day but it sure adds up folding.

[RollinLower]

43 minutes ago, Gorgon said:

Yes, GPUs like CPUs these days are designed to run near the top end of the voltage-frequency curve just a “titch” below the point of instability to account for variations in quality. The result is they are horribly inefficient when run at stock settings.

 

it’s not an issue when gaming for a few hours a day but it sure adds up folding.

yup, my F@H box has limits set to 65% and 70% on both cards, PPD loss is minimal, but the difference it makes in temperature is astonishing!

[rkv_2401]

4 hours ago, jctappel67 said:

Tested this while running WU 18201. Cut cards power limit to 65% (~200W). It's running about 8 C cooler with a loss of only 300k ppd. I'll take it if the darned thing doesn't make my room into an oven anymore.

I use a custom profile with a locked frequency for Folding, playing video games and everything else demanding for each of my cards. With a well-tuned undervolt specific to that frequency, that's as efficient as that card is ever going to be!

[Gorgon]

7 hours ago, rkv_2401 said:

I use a custom profile with a locked frequency for Folding, playing video games and everything else demanding for each of my cards. With a well-tuned undervolt specific to that frequency, that's as efficient as that card is ever going to be!

Now if the only there was under-volting support on Linux.

 

Honestly though I've found over the years the simplest solution is the easiest for long term stability. I was adjusting the PBO limits on CPUs but I recently figured out how to just limit the maximum frequency which works much better. Mind you I don't game on any of these systems so they're dedicated just to distributed computing so it's likely easier.

[Gorgon]

And here's some p18041 (only 170,000 Atoms) efficiency readings on a RTX2060 Super:

That R^2 value of 1.0 is impressive. I messed up the 18042 data by using the "set" Power-Limit rather than the "observed" Power-Limit from nvidia-smi.

[rkv_2401]

1 hour ago, Gorgon said:

Now if the only there was under-volting support on Linux.

 

Honestly though I've found over the years the simplest solution is the easiest for long term stability. I was adjusting the PBO limits on CPUs but I recently figured out how to just limit the maximum frequency which works much better. Mind you I don't game on any of these systems so they're dedicated just to distributed computing so it's likely easier.

Yeah, I agree that your way gets 80% of the result with far less effort, time (and OS specificity, seemingly - did not know you couldn't undervolt on Linux). I had 2 cards and a lot of time to kill - I don't expect my method to be feasible with someone like you who's managing 8 - 10 systems, some with multiple cards. Hell, I probably wouldn't do it again myself, considering my schedule now. I put off installing another SSD in my laptop for like 3 days. But back then, it gave me a bit of happiness to know my card was as well tuned as I can get it to be for a certain task - same way some people feel when messing with their cars, I guess..

[GOTSpectrum]

come own up, we've all done it. 

 

Accidentally didn't set the user for days at a time after a new headless install...

 

 

[Gorgon]

A new member has joined my fold.

Of course I also bought more RAM and another 750W power supply so I'll be busy tearing apart my rigs this weekend and re-building them with Ubuntu 20.04.

[RollinLower]

i need some help lads. 

i recently reinstalled FAH on my dedicated box, but it seems like it;s not picking up my username and passkey. I've been running for 2 days on all CPU and GPU's, but my graph on EOC shows no points.

what gives?

my config file:

 

[Lightwreather]

8 minutes ago, RollinLower said:

-snip-

Might want to blur the passkey.

[RollinLower]

4 minutes ago, J-from-Nucleon said:

Might want to blur the passkey.

didn't think there where people out here low enough to go stealing FAH users, but you never know i guess. it's not like it;s a bank PIN or something

[RollinLower]

so turns out the client wasn't actually looking to the config.xml from /etc/fahclient/config.xml, but from another config file under root.

after setting that one up with the proper info i seem to be in business!

 

freakin Linux....

 

[Gorgon]

4 hours ago, RollinLower said:

so turns out the client wasn't actually looking to the config.xml from /etc/fahclient/config.xml, but from another config file under root.

after setting that one up with the proper info i seem to be in business! ...

Good to know. Which Distribution?

[Captainmarino]

22 hours ago, GOTSpectrum said:

come own up, we've all done it. 

 

Accidentally didn't set the user for days at a time after a new headless install...

 

 

I'm actually proud to say that each and every single one of my 9832 WUs has been for LTT

[RollinLower]

23 hours ago, Gorgon said:

Good to know. Which Distribution?

This is on ubuntu server 22.04.