Low Power Folding - Reduce Your Power Draw & Heat

[BWG]

A while ago, I had started a movement to try Low Power Folding on ocn. Some people jumped on board, and we found it was helpful to beating the Summer Heat and limiting the power bill costs. Well, some advancements have been made to GPUs since then, so I think it's time to revisit the subject. In the coming weeks, I will be publishing data showing the benefits of low power folding and the detriments.

 

For a general and immediate understanding of how this works, I'm going to use examples of mining on a GPU because you see instant measurable results. One thing to keep in mind is mining stresses memory and folding stresses core. They're completely different animals, but the process is the same with lowering the power limit. But, for now, mining will be the example used and in the coming weeks, folding will be added.

 

Our test GPU will be a beautiful EVGA 3080ti FTW3 Ultra Hydrocopper. 

 

All you really need to do is install your favorite overclocking utility such as Afterburner or Precision. The screen you see below is Precision X1. I use it because of how many EVGA GPUs I have, and the RGB is controlled within the program.

 

 

 

If you want to run stock clocks, then all you need to do is push the power slider down, or type in the % of the power limit you want to use. This works differently on each GPU. Just because 1 3080ti works best at a hypothetical number of 75% doesn't mean another will be optimal at the same limit. You have to test your card individually. 

 

If you're overclocking, you can still benefit from doing so even at a lower power limit. So, I suggest you test to find your stable oc at full power, or even over the power limit of 100%, and then move the power limit down gradually to test. You can also reduce the memory clock with folding kinda the opposite way some people do on RTX cards with the core clock for mining. Doing so can lower the stress on the non-critical component (folding - memory / mining - core clock). 

 

The quick example for today of doing this with mining is to mine Autolykos (Ergo) since an 80ti has LHR (Limited Has Rate). For mining, I'm overclocking the memory speed. For folding, you overclock the core speed. I currently run a memory overclock of +1370 MHz for mining Autolykos. At 100% power, the GPU consumes 331 watts and produces a steady hashrate of 266 MH/s. If I reduce the core clock by 150 MHz, the power draw decreases to 314 watts, but the hash rate climbs to 271 MH/s. This is a result of lowering the stress on the core, improving temps, and efficiency. I can easily display all of this data on 1 screen for mining within seconds, so that's why I wanted to do this right now with mining.

 

 

The goal in mining is to lower the power limit slowly until you see the hash rate being affected. Sometimes, the production increases due to lower temps, or it can decrease the hash rate drastically. Either way, reducing the power limit reduces the amount of power you use. The Eff/Watt measurement helps miners see if the decrease results in more earnings or not. It helps figure out the balance between the 2 and which is better.

 

I've already tested this GPU to find it's sweet spot. It's most efficient at 75%. If I go 1% less, the hash rate drops drastically to the 150 MH/s range and bounces around. But, at 75%, I lose nothing and gain my wasted power cost back. This is what I am going to test for on folding in the coming weeks. How much PPD do you lose? Where's the sweet spot, etc. As you can see here, I'm now using 298 watts and getting 272 MH/s.

 

 

So, how will the results look for folding? Folding uses a ton more power than mining. The wattage decrease will be far more drastic, but all of those details will be provided in the coming weeks. I already know my overclock potential on this card. I can run +180 to the core, +100mv, and at 113% power. I ran those clocks on it for the 2 day foldathon which is monitored by HFM. The results for the work it did are below, and this will be compared to the power limit reductions to see what we're gaining/losing. I will have to re-measure the power draw though with this overclock, and will be updating this thread as I go.

 

 

Alright, so here's where we are on the power draw running folding with the OC used to obtain the above PPD. The actual clock speed is 2160 MHz.

 

 

It fluctuates between 401 and 404 watts. I'm folding right now. I will begin to adjust the power in 5% increments at first just to see if I notice a significant drop in TPF, and then gauge from there where the sweet spot might be. Folding might be a bit of a challenge, but I think the key here is this card s not a display GPU, so I shouldn't see usage spikes up/down. 

 

95%

p18004

 

p17804

p17713 - Core jumped around on this project. GPU-Z says PerfCap Reason Pwr.

p18202 - Core jumping, pwr (CJP)

p18005

p17804

p13456

 

90%

85%

80%

75%

70%

65%

60%

55%

50%

p13456 CJP

p18202 CJP

Edited July 29, 2021 by BWG

[BWG]

There's a flaw in today's earlier data, so collection on 95% will run through the end of tomorrow, but thus far, the tpf is 53 seconds on p18004 which matches my tpf from above. The core clock is steady at 2160.

[tictoc]

This is a great topic.  

Having mostly ran AMD GPUs for the past decade, I'm pretty used to the notion of undervolting+overclocking (especially in the last few generations) to achieve equal performance at lower power draw.  Efficiency is king, especially for systems that are going to run 24/7.

[BWG]

17 hours ago, tictoc said:

This is a great topic.  

Having mostly ran AMD GPUs for the past decade, I'm pretty used to the notion of undervolting+overclocking (especially in the last few generations) to achieve equal performance at lower power draw.  Efficiency is king, especially for systems that are going to run 24/7.

 

Yeah, some people that care about heat, draw, longevity, and efficiency should find everything relevant, I hope.

 

I'm going to just go ahead and jump down to 50% tomorrow just to record the drastic change everyone is probably hoping to see sooner rather than a gradual decrease of 5% incrementally. I hope the crowd goes wild!

[Bastiaan_NL]

Really interesting @BWG

I'm running my 3090 @ 95% for a few days now, and it doesn't seem to care at all. I've even seen up to 8 mil ppd on P18202 which is close to the max I've seen on that unit. Besides that it runs a few degrees cooler too.
I wouldn't really care about this if I'm watercooling, but on air it sure does help a lot!

[BWG]

I keep pulling the same work today at 50%, so I might have to run 50% again tomorrow. OP gets updated as I go with watts at least. Under 200, but it's taking a huge PPD hit, I think.

[damric]

Bump for @mx500torid

 

This is very relevant with people's power bills exploding during this heat wave. This helps especially if you have a high tier, high TDP graphics card, and even more so if you have multiple.

 

For the last several folding contests I was running my RX 6900XT at less than half power but getting about 80% of the ppd. Not heating up the office during the summer!

 

 

 

[mx500torid]

Thanks to damric for the idea. My 2080 went from 225 watts to 125-130 and ppd did not fall that much. Tower was pulling 220 watts. 

 

 

 

Edited July 27, 2023 by mx500torid
add more info

[damric]

6 minutes ago, mx500torid said:

Thanks to damric for the idea. My 2080 went from 225 watts to 125-130 and ppd did not fall that much.

 

 

 

Play around with lowering the memory clocks if you can. Folding isn't very bandwidth intensive compared to gaming or mining, so finding the spot where you can drop the RAM clock and keep the same performance will also save some watts.

[ENTERPRISE]

Fantastic data @BWG

 

Reducing the power used by the GPU's when folding/mining has multiple advantages as you state. Keeping the bills and heat low not only benefits the individual but decreasing the power usage lends to more green and responsible energy usage. It is a win across the board