Graphic Cards Do you undervolt + slightly downclock your GPU

Do you undervolt + slightly downclock your GPU for

  • Yes (Tune for power and thermal efficiency)

    Votes: 14 58.3%
  • No (Stock OOB experience)

    Votes: 5 20.8%
  • POWER!! (OC the crap out of it. Grid collapse and rising sea levels be damned)

    Votes: 5 20.8%

  • Total voters
    24

Psycho_McCrazy

Disciple
I recently acquired a 1080Ti
Awesome card, MSI twinfrozr OC model. At its oc'd clock settings, i've seen it draw up to 280W, but that results in high thermals and then clocks drop to compensate.

I have taken to undervolting + slight downclocking (still way above nVidia's official boost clocks). For me, this results in very stable clocks, much better thermals and power consumption down by a massive 30%. Card is extremely happy at 180 odd watts.

Was wondering how many of us do the same with their GPUs
 
yes, last 15-30% seems to be very inefficient for 3080. Even more wasteful for 4090 from what i read.

3080 Zotac loses about 3-4% performance at 85% max power and additional 7-9% for next 15%. Fans/thermals are much better with lower PL.
So on Linux, where i cannot undervolt, i set max power limit at 225w ( about 70%).
And on windows, i undervolt + downclock slightly - this is actually faster than stock at 100% PL. But then i use 85% PL by default for very minor performance loss vs stock. Happy with this so far.
 
I run my 3070 undervolted and sight downclocked. The temperature difference and power consumption is significant. I don't recall the exact wattage number but roughly it was around 40 watts drop and 10 degrees drop while giving almost similar performance in games (well within the error margin) and the performance even increased in some benchmark because there was no thermal throttling.
 
I run my 3070 undervolted and sight downclocked. The temperature difference and power consumption is significant. I don't recall the exact wattage number but roughly it was around 40 watts drop and 10 degrees drop while giving almost similar performance in games (well within the error margin) and the performance even increased in some benchmark because there was no thermal throttling.
What clocks are you running? Did you downclock it using the graph on afterburner? Turning down and getting better performance just doesn't make sense though...
 
Back in the day I used to undervolt and reduce clock speeds slights of my GTX 1060 In order to reduce heat and power consumption. There hardly couple of fps lost but the card used to run smoother and there were no frame drops. People focus so much on overclocking that for folks who prefer peace and calm the concept of undervolting is lost.
 
What clocks are you running? Did you downclock it using the graph on afterburner? Turning down and getting better performance just doesn't make sense though...
I don't remember it exactly as I did it months ago. I followed a YouTube video specific for my card. (Asus Turbo 3070 v2)

Earlier it was thermally throttling and hence the massive dips in performance on extended benchmarks. This is what I assumed.
What clocks are you running? Did you downclock it using the graph on afterburner? Turning down and getting better performance just doesn't make sense though...
I don't remember it exactly as I did it months ago. I followed a YouTube video specific for my card. (Asus Turbo 3070 v2)

Earlier it was thermally throttling and hence the massive dips in performance on extended benchmarks. This is what I assumed.
 
I run my 3060ti at stock as the fan itself hasn't spun for almost a year now/ no temp issues here and utilization is barely 15-20%. No time for gaming.. So no tuning or fiddling required in my case.
1080 3080 are power hungry cards for which it makes sense to tune them for efficiency.
 
Have a laptop so undervolting is kinda necessary ¯\_(ツ)_/¯

Undervolted my Helios 300 w/ 105W RTX 3060 to 1876MHz @ 881mV (compared to stock boost of about 1880Mhz boost at about 1000mV). Attaching graph of the UV curve done via MSI Afterburner. Reduced temps by about 6C with a very slight performance increase.
The CPU (10870H) is also undervolted to -201.2mV on the Core, -92.8 on the Cache, -49.8 on the System Agent, and -42 for the Intel GPU and iGPU Unslice. It can maintain an all-core turbo of 4GHz for stuff like video rendering and gaming without going beyond 90C (basically everything but synthetic benchmarks, for my use cases), compared to being at 3.6GHz all core and 95C.
Doing both CPU and GPU undervolting has reduced temps while gaming by about 8-10C.

Overall, very happy with the results and would wholeheartedly recommend undervolting. This setup has been serving me very well for nearly 3 years now :D
Loss of undervolting was a big blow to 11th gen and beyond laptops; hopefully it makes a comeback soon.
 

Attachments

  • 1677741759582.png
    1677741759582.png
    240.5 KB · Views: 53
I actually tried testing undervolt settings when I got my GPU because everybody said it's the next best thing since sliced bread, and here's what I found from my sample size of 1:
After seeing what tutorials were suggesting, I thought to myself, "Well, this isn't undervolting, this is undervolting +OCing" since they were figuring out a stable boost clock on stock curve and trying to achieve that stable clock throughout their gaming session, which is not a bad thing actually but the stock fan curve is what throws me off(Still would recommend UV + OC over nothing).

I decided to see whether having a "stable" clockspeed vs having it boost up and down would really make that much of a difference (In theory it'd make sense to have a stable clock instead of having your card keep boosting and throttling over and over right?) After looking at my saturation point and getting a stable clock of 1950(?), I checked power usage and temps and decided to dial in both the curve constant clockspeed method and did a benchmark run.

After this, for the boost benchmark, I checked how much power the card was consuming and what temperature it was hitting, I changed the power target to what the card was consuming at constant clock speed and ran the benchmark again and all the variables(temps,power draw) were about the same as with a constant clock speed. The only difference? I actually managed to get a higher score on it. This sort of was an answer for the question I had before testing that whether a humongous company like NVidia would mess up their GPU boosting behaviour so bad? Turns out I was getting 1945-1985 clocks throughout the run. And the boosting behaviour was actually working as expected.

Conclusion:From my testing it was clear that people should be targeting fan noise and power draw (&temperature) instead looking at stock/boost clocks and just set that(comfortable noise level/ acceptable heat output) using the power slider on Afterburner then just push the GPU and memory clocks as high as possible while being stable(Also please for the love of god test them in-game).

P.S : I am not an NVidia engineer nor der8auer himself but this has been a much cut down version of my findings through my testing of basically a week straight and noting down the performance of my card.
1677767605735.png

I wanted to actually transcribe everything I found out but meh sometimes condensing information is a good thing.

TL;DR : Atleast UV + OC your card, but focus should be more toward hitting a noise level or thermal output when it comes to undervolting.(Constant clock speed vs auto boosting shouldn't really matter as much)
 
After this, for the boost benchmark, I checked how much power the card was consuming and what temperature it was hitting, I changed the power target to what the card was consuming at constant clock speed and ran the benchmark again and all the variables(temps,power draw) were about the same as with a constant clock speed. The only difference? I actually managed to get a higher score on it. This sort of was an answer for the question I had before testing that whether a humongous company like NVidia would mess up their GPU boosting behaviour so bad? Turns out I was getting 1945-1985 clocks throughout the run. And the boosting behaviour was actually working as expected.
This is expected behaviour for desktop cards and how nVidia's boost behaviour works keeping in mind the thermal and power constraints of the device. Just a UV is often enough for desktops to reach essentially the best boost clocks since the desktop cards have much more thermal headroom than laptops. But for laptops, UV+OC matters, since the voltage and clock table is much more constrained by thermal limitations. In my case, the card was going no higher than about 1700MHz (which is already boosted from the voltage table target of about 1640MHz) without also overclocking the undervolt curve.
 
Good to see that most of us tinker with the settings and try to figure out what we feel is better for our specific use case.
The work put in by Psycharge seems immense, especially compared to mine where I just set a voltage/clock point in afterburner and verify its stability by playing games...
The target of-course as he suggests is a significantly reduced thermal / power footprint, and consequently better acoustics (much lower fan speed).
 
Back
Top