Folding@home Beyond3D_Team - Summer 2023 Call for NVidia RTX users

I'd probably fold a lot more if I could easily set the folding client to limit GPU usage/power. Yes, I know I can use MSI Afterburner or similar, but that also requires remembering to do it, and remembering to set it back.
I've been using Asus GPU tweak III for years and there is a feature called Profile Connect that offers automatic profile swapping where you might want specific tuning (clock/voltage/fan) settings when a game or app loads. It sounds similar to what is offered in the AMD Adrenalin software.

"Profile Connect allows settings to be saved and linked to specific applications. Overclock settings, fan curves, and even customized OSD settings can all be saved and automatically loaded whenever a target app runs."
 
The earlier Asus GPU Tweak versions were kinda crap, I tried using them and gave up and went back to MSI.

However it's been literally years since I've tried, so it's worth trying again. Thanks for reminding us all @pharma ! I could potentially see using this for even slower speed undervolt settings for my older games library items like Elder Scrolls and Fallouts and the like.
 
We're now at #348 with 6 billion points - at the current rate of ~30M points per day, that's 1.5 months before we earn 7.1 billion points and enter Top 300:


Rank
Project
Team
Name
Points
24hr Avg
Points
Total
Points
Diff
Gain
Daily
Date
Overtake
300​
TogetherWeArePowerful
5,124,918​
7,148,263,927​
-1,138,089,054​
24,595,019​
03.08.24, 7am / 1.5 Months​
348​
Beyond3D_Team
29,719,937​
6,010,174,873​
0​
0​
--​

Please join Beyond3D_Team to help us climb faster - and soon you will find it incredibly easy to justify an Nvidia RTX 4090 card as a high-efficiency upgrade for your heating furnace! 🔥
 
Last edited:
No update on my pending Asus TUF 4090 OG OC order other than it's "processing" since Sunday, bleh. I also have an opportunity to pick up a 2060 for super cheap, so I'm considering it... Also my old Core i7 3930k sits on the venerable x79 chipset, which provides 48 lanes of PCIe 3.0 via three full size 16x connectors. Folding doesn't really need the PCIe bandwidth, however physically it gives me the ability to load up a few cards for a dedicated folding rig.

On a lark, I loaded F@H on my laptop just to see, and discovered (among other things) there is some rudimentary support for Intel IGPs? Went looking in the LARS website and apparently they're good for about 30k PPD, which is laughably tiny. Anyway, the 1070MaxQ on that laptop churns out around 900K PPD at 1685MHz and about 65W total system power, which isn't terrible. I deleted the CPU crunching slot as it moved the temps from about 70*C to over 85*C where everything started to throttle hard. All that extra heat was only worth about 40K PPD estimated, while the whole laptop power consumption shot to over 100W. Not worth it...
 
Last edited:
You know it!

To make it worse, neither the x79 chipset nor the 3930k were ever sold as anything close to "power efficient"; I think that rig had a 150W idle power draw :D Not sure what I'll do with a 600W space heater in the corner of my office during the hot southern summers here in Tennessee. Guess I'll cross that bridge when I come to it.

My 4090 has shipped and should be here on Monday. I'm excited to start undervolt testing soon...
 
The 4090 arrived yesterday, as expected. Good grief the box is huge and the card is huge! One reason why I bought this specific model is its generally one of the shortest factory-OC 4090 models, and yet it's still a full inch longer than my EVGA 3080Ti FTW3 card. I had a hell of a time cramming it into my uATX Silverstone case, which already had a 280MM radiator shoved into it. Anyway, after literally taking the case apart and then reassembling, I was able to get it fitted.

Now I'm working through the undervolt, using Cyberpunk 2077 raytracing as my stability tester, as I discovered it died far sooner than any other graphics stability testing software I've used before. Preliminary results at standard 100% power limit are:
0.880v at 2525MHz
0.900v at 2580MHz
0.925v at 2650MHz
0.950v at 2715MHz (I'm still testing this)

My PPD is lookin' pretty swanky :) Just wait until I get this 3080Ti moved over to my new x79 / 3930k folding rig!
 
Wooooooooooooboy! A tale of woe and redemption for anyone considering a Linux folding rig.

I became so accustomed to my whopper undervolt + overclock on my 3080Ti in Windows, I didn't stop to think about how it performs stock. Power consumption of that whole box, at the wall, was a whopping 835 watts! This is 100% GPU folding only, the super-thirsty 3930k is not in the picture at this time (minus whatever it does to feed the pair of GPUs.)

A tiny bit of investigation later, Nvidia-smi shows the 3080Ti gluttonously inhaling all 400W of the driver power limit with some weaksauce 1580MHz clocks! As for the 2060, it's "new" to me and I'm unsure of undervolt-ability, yet it was also sitting at the 175W driver power limit around 1700Mhz.

For those who don't know, NVIDIA's Linux drivers do not expose voltage controls. I hear what some of you might be suggesting -- "Alby, just turn down the power limits and the card will figure it out." And it kinda does, except the performance drop is not commensurate with the power limit drop IMO. Sure, I can configure a power limit (nvidia-smi -i 0 -pl 275) of 275 watts, but the performance is honestly bad compared to what I know this card can do.

Linux drivers do expose overclock offsets, and when combined with hard-limiting the max clock frequencies, you can thus achieve the same undervolt result. This is because the overclock offset simply moves the top clock speed for each voltage bin... So said another way, with a clock offset of +255MHz and a clock limit of 1785MHz, it settles right back to my preferred 800mv voltage bin. And wow is it better; performance is up ~15% and power draw is down 40%, as compared to just letting it run at stock. Good freakin' grief.

I'm now futzing with the 2060 card as I'm sure there's some extra to be had there, too. Nevertheless, the whole rig is now down to 510W at the wall, instead of 830W. Some of this is due to fans running slower, some of it is powersupply overhead too.

For those of you who fold, you should REALLY be looking at undervolting.
 
Thank you to everyone who continues to fold, we're less than two months away from breaking into the top 250!

Anyone else want to contribute a few cycles to the cause? You can use the Folding After Dark browser plug-in to allow your machine to fold at night while you're asleep, lessening your winter heating bills :)
 
@pharma per your suggestion, I tried out Asus GPU Tweak 3 today for a few hours of testing and a few million less PPD! Here are my thoughts:
  • Pro: It's better than I remember, in terms of UI and monitoring and just general presentation.
  • Pro: For the generic "let's overclock to the max" stuff, it seems to work just as well as MSI Afterburner.
  • Pro: I tried creating app profiles and it worked well too. Did a minimal clock profile for Oblivion and Skyrim, did a max overclock profile for Cyberpunk, did stock clocks for MS Flight Simulator -- it all worked as expected. Creating a profile for Folding didn't work because the binary is always loaded, albeit sometimes it's on pause... I think I'd need to create profiles for each folding "core" to make it work correctly, however those cores are often updated and I suspect a new version of the FAH core would break the overclock profile.
  • Con: The only way to access any quantity of curve editing is to first permit it to run a full OC Scanner session, which takes as long as 20 minutes.
  • Con: Once I was able to get to the curve editor, actually creating a proper undervolt curve by selecting a point and flattening literally every point to the right at the same height doesn't work properly. There's a specific button for "flatten to lower value" in the editor, and it draws a flat line like you expect. However, as soon as you apply it, a curve shows up starting at an arbitrary point midway through the flat line. The result: the card simply hops from your desired undervolt alllllll the way over to some crazy voltage at an incrementally higher speed.
  • Con: No amount of fidgeting with the curve editor allowed me to get a proper undervolt to work, either due to the insertion of a curve without my asking for it, or because the 4090 doesn't respect any voltage curve settings below 870mV.

After working on it for hours, I finally gave up and went back to MSI Afterburner. Humorously enough, my experience with GPU Tweak did show me a new trick and also reminded me I found my max stable clocks / temp / volt combo for my 4090. As such, I ended up building a whole new set of clock profiles in MSI which gave me a far better set of power vs performance options to choose from.

Setting 1: 2205 MHz GPU at 870mV, 5001 MHz MEM. Apparently the 870mV voltage setting is not only the lowest accepted programmable voltage level, it also places the card in the P2 power state which forces memory clocks to half speed. In this profile the card runs at ~240W absolute max and folding comes in below 190W. This setting is overkill for >90% of the games I play.

Setting 2: 2505 MHz GPU at 875mV, 10251 MHz (stock speed) MEM. The card doesn't seem to exceed 300W no matter what I throw at it, and it's great for Cyberpunk 2077 Ultra and pathraced at 3440x1440 100Hz with DLSS Quality and FG enabled. Folding in this profile draws ~215W for around 23M PPD on a good WU.

Setting 3: 2625MHz GPU at 915mV, 10751 MHz MEM. This is preferred folding setting, averaging ~245W and pumping out just shy of 25M PPD on a good WU. Fans hover around 75% while temps say below 70*C. Playing CP2077 on this profile will chew through ~330W and sometimes more, unfortunately the case fans start picking up speed so it's not as quiet as the slower settings.

Setting 4: 2745 MHz GPU at 965mV, 11251 MHz MEM. This is a bit of a burner, it's good for e-peen gloating and warming up a room. Folding at this speed will absolutely exceed 280W and I've seen PPD numbers slightly exceeding 26M, but it's not worth the power draw and heat. Playing CP2077 will get it to 400W...

Setting 5: 2865 MHz at 1025mV, 11925 MHz MEM and power limit at 111%. The card seems stable here over the course of a few hours of CP2077 and some folding, but it occasionally bumps into the firmware's 500W max power limit while Cyberpunking. Folding gets a tiny performance bump to 27M PPD on a really good WU, yet draws ~355W on average. CP2077 turns the whole box into a convection oven while riding that 500W power limit, while FanControl shoves every fan in the case to 100% in protest of crazy heat coming off the card. I did this mostly "just to see", otherwise it's a pointless profile.

A fun exercise, and learning about that P2 power state to cut the memory rate in half was actually quite useful for throttling down the card for simpler games.
 
Last edited:
Bump!

We are at position 306. which means we should break into the top 300 early next week!

Come fold with us for lots of great causes!!
 
We're now at #300 with 7.34 billion points - at the current rate of ~55M points per day, it's ~3 months before we earn another 5.5 billion points and sneak into Top 200 - and merely a year before we earn ~28 billion points and enter the Top 100! :runaway:

Rank
Project
Team
Name
Points
24hr Avg
Points
Total
Points
Diff
Gain
Daily
Date
Overtake
100​
Jump Trading
0​
28,638,814,760​
-21,300,901,041​
57,279,284​
02.25.25, 12am / 1 Year​
200​
Salesforce.com
5,124,918​
12,846,108,997​
-5,508,195,278​
57,080,703​
05.25.24, 2pm / 3.2 Months​
300​
Beyond3D_Team
57,279,284​
7,337,913,719​
0​
0​
--​

So please join Beyond3D_Team and help us make it before the end of the current folding heating season! 🔥
 
Last edited:
FYI, NVidia has released a new driver control panel, aptly named 'NVidia App', which consolidates per-game settings from previous NVidia Control Panel and GeForce Experience applications:


They promise to implement additional overclocking controls in the future, though it's not clear if they will include manual tweaking of GPU/memory clocks and voltages, similar to AMD Software Adrenalin control panel and various RivaTuner-derived tools like MSI Afterburner, ASUS GPUTweak, and EVGA Precision X1...

The NVIDIA app beta is a first step in our journey to modernize and unify the NVIDIA Control Panel, GeForce Experience, and RTX Experience apps.... Featuring a unified GPU control center, NVIDIA app allows fine-tuning of game and driver settings from a single place, while introducing a redesigned in-game overlay for convenient access to powerful gameplay recording tools, performance monitoring overlays, and game enhancing filters, including innovative new AI-powered filters for GeForce RTX users....

Moving forward, we’ll be integrating the remaining features from the NVIDIA Control Panel... Additionally, we'll be adding several attributes from GeForce Experience and RTX Experience, such as GPU overclocking and driver roll-back.... On our beta roadmap, we plan to offer AV1 support for Shadowplay, additional DLSS controls, extra overclocking options, and more.

PS. And unlike GeForce Experience, the NVidia App doesn't require you to login into their web services in order to actually work! Only took them 8 years to respond to never-ending user complaints about this 'feature'...
 
Last edited:
They promise to implement additional overclocking controls in the future, though it's not clear if thet will include manual tweaking of GPU/memory clocks and voltages, similar to AMD Software Adrenalin control panel and various [URL='https://en.wikipedia.org/wiki/RivaTuner']RivaTuner-derived tools like MSI Afterburner and ASUS GPUTweak...
Since EVGA no longer produces gpus Nvidia should save development effort and purchase the EVGA Precision X1 tool and modify accordingly.
It basically only works with RTX and GTX Nvidia cards anyway.
 
Not sure who needs to know this, but it seems the 4070 Super is now the add-on dGPU leader for best Folding PPD per kWh and arguably the best MSRP per PPD of any new card. $589 for 12MPPD is pretty damned solid.


As soon as I see a quality 4070 Super go on discount, I'm gonna buy two and swap the 3080Ti + 2060 out of the dedicated.Linux folding rig.
 
Last edited:
The recent additions to the thread finally reminded me to get unlazy and set this all up.
Added my 3080 and 3090 to the mix. :)

Getting to the part of the year in the Northern hemisphere where I don't need heating during the day, but I've also been meaning to add some software defined heaters to my home automation via Home Assistant.
Why have it trigger an actual heater when my house gets chilly when it can just start Folding instead?
 
I left town this week and noticed my 4090 rig has stopped folding. I'm 100% sure it's a stupid Windows patch that force-rebooted my rig. So I'm 25MPPD down until I get back on Sunday evening 😢
 
Added my 3080 and 3090 to the mix
Why ... an actual heater when my house gets chilly when it can just start Folding instead?

Great - you should be making ~18M PPD with your energy-efficient hi-tec folding heating rig! 🔥

my 4090 rig has stopped folding
So I'm 25MPPD down until I get back on Sunday evening

And your room is probably 30 ° cooler too... ❄️ you should really consider another RTX 4090 as a backup heater! 🔥
 
Last edited:
Back
Top