Rtx 3000 series undervolt discussion

So, I've just got my 3070, but can't see its voltage in MSI Afterburner. Is that because I'm using 4.6.2 instead of 4.6.3 Beta 2?
Besides, I did one run of Heaven at stock and noticed the GPU was always power-limited. Is that normal?
Cards will power limit normally, yes. I don't think the 30 series cards have voltage control but I'm sure someone can correct me.
 
Cards will power limit normally, yes. I don't think the 30 series cards have voltage control but I'm sure someone can correct me.
I believe they have to update Afterburner to the latest version to support the RTX 30xx series cards. Also, I think to view voltage in the monitoring graphs, you have to enable voltage monitoring in the settings.
 
I believe they have to update Afterburner to the latest version to support the RTX 30xx series cards. Also, I think to view voltage in the monitoring graphs, you have to enable voltage monitoring in the settings.
Thanks for the info. I usually use gpuz for monitoring and logs so I never really felt the need to set it up in afterburner.
 
So I did some runs in Heaven and these are the results:

1965MHz @ 1075mV @ 100% PL - Stock GPU
2040MHz @ 1000mV @ 104% PL - Stable OC; min fps 35.7; avg fps 110.6; score 2786
2040MHz @ 950mV @ 104% PL - Unstable OC
2040MHz @ 975mV @ 104% PL - Stable OC; min fps 35.7; avg fps; 109.6; score 2760

Do you think it would be fine keeping the GPU at 975mV? I got slightly lower avg fps.
 
So I did some runs in Heaven and these are the results:

1965MHz @ 1075mV @ 100% PL - Stock GPU
2040MHz @ 1000mV @ 104% PL - Stable OC; min fps 35.7; avg fps 110.6; score 2786
2040MHz @ 950mV @ 104% PL - Unstable OC
2040MHz @ 975mV @ 104% PL - Stable OC; min fps 35.7; avg fps; 109.6; score 2760

Do you think it would be fine keeping the GPU at 975mV? I got slightly lower avg fps.
You will probably find instability in other games or applications if you are that close to the edge. Might want to set a lower clock to 2025 or bump it to 1000mv for peace of mind.
 
You will probably find instability in other games or applications if you are that close to the edge. Might want to set a lower clock to 2025 or bump it to 1000mv for peace of mind.
I'm going to try TimeSpy to see if there's any difference in stability with DX12. Anyway, do you think it would be bad to use voltages that aren't 25 multiplies? Like 987mV for example. Even at 1000mV GPU was very cool peaking at around 60°C and drawing around 200-220W (same as NVIDIA spec for 3070 GPU).
 
I'm going to try TimeSpy to see if there's any difference in stability with DX12. Anyway, do you think it would be bad to use voltages that aren't 25 multiplies? Like 987mV for example. Even at 1000mV GPU was very cool peaking at around 60°C and drawing around 200-220W (same as NVIDIA spec for 3070 GPU).
No problem using any voltage as long as you are stepping the frequency by 15.
 
Stepping frequency by 15? Btw, I forgot to mention that when setting 975mV as frequency, it was actually running at 967mV (dunno why).
Frequency offset should be a multiple of 15. If it was pulling less voltage than what you set, either you didn't use an offset that's a multiple of 15 and therefore it rounded down (to the frequency set at 967mv) or you were vrel/pwr limited. Gpuz will tell the full story.
 
Frequency offset should be a multiple of 15. If it was pulling less voltage than what you set, either you didn't use an offset that's a multiple of 15 and therefore it rounded down (to the frequency set at 967mv) or you were vrel/pwr limited. Gpuz will tell the full story.
GPU-Z says "Idle" in PerfCap Reason while running the benchmark, is that normal?
 
GPU-Z says "Idle" in PerfCap Reason while running the benchmark, is that normal?
No, that's not normal. Typically you'd see power limiting.

ed- I imagine there may be different types of values for perfcap depending on the curve, but it still seems odd that idle would be possible when the GPU is boosting.
 
No, that's not normal. Typically you'd see power limiting.

ed- I imagine there may be different types of values for perfcap depending on the curve, but it still seems odd that idle would be possible when the GPU is boosting.
I'm fairly certain if you are at a low enough load and you have a cap on frequency at a lower than peak voltage like he does, you can see idle as perfcap.
 
This is my GPU after running TimeSpy:

abc.png
 
I still find that odd. His comment above the screen shot says it's after running the benchmark, not during. I can't imagine it says it's idle when the benchmark is running. What's it say during the benchmark?
 
It looks like 2040MHz @ 981mV is perfectly stable, that's with +150 shift compared to the stock curve. Even though frequency was at 2025MHz during most of the test.
 
I still find that odd. His comment above the screen shot says it's after running the benchmark, not during. I can't imagine it says it's idle when the benchmark is running. What's it say during the benchmark?
I'm pretty sure thats normal. The gray perfcap is idle which you can see is while the GPU is under load.
 
So what does the "power limit" in Afterburner do? How come it's under the "fan" setting, and how come no other undervolting instructions ever mention it?
 
My ASUS TUF runs around 2000-2055 MHz without doing any undervolting. Also runs at 65C or below. What additional benefit am I looking at by under volting?
 
My ASUS TUF runs around 2000-2055 MHz without doing any undervolting. Also runs at 65C or below. What additional benefit am I looking at by under volting?
Less time in your favorite game :D

Since each card can be different, only way to know is to try it out. I may when I get a chance. If 25w-60w is important to you and maybe a 2% gain, I probably would not notice but maybe fun between game breaks.
 
Watts are definitely not important to me. Higher sustained clocks and less heat are. If I can undervolt and run it at 2070-2100 sustained clocks it might be worth a shot.
 
Watts are definitely not important to me. Higher sustained clocks and less heat are. If I can undervolt and run it at 2070-2100 sustained clocks it might be worth a shot.
watts = heat. 2100 sustained I don't see happening, if by sustained you mean average, but you should definitely be able to lose 30 watts of heat without losing performance, is my guess.
 
gpucurve.jpg

This is on a 3080. I didn't do it step-wise like the OP because I wasn't sure if it's better or not, and doing it flat like this seemed to be the most shown in undervolting tutorials so I went with that.
I got slightly better score in 3DMark Time Spy with this than stock and only 10W less used average but 30-40W *maximum load* used, bit less fan speed.

Anyway, does this curve seem ok?
90% power limit, 875mv @ 1935mhz. Arbitrarily picked 875. Crashed on Time Spy at 1950, lowered 15mhz and ran BMI and Time Spy successfully and called it a day for now.
Anything I should change about it?

ps is getting frame-time spikes/micro pauses during BMI stress test normal? It stutters in the same sections no matter if stock or undervolted. For example right when the bullet through the glass bottle scene shows is a stutter. Wondering if it happens to others? All other games (and 3DMark) never had stutters for me, everything has been butter smooth

edit: also passes 45 minutes of Control gameplay. I was switching between stock and the undervolt and I would get a couple frames extra undervolted, as a matter of fact.

Was a relatively easy process despite me not knowing exactly what I'm doing and the intricacies involved. I confirmed with GPU-Z my max voltage matched to the undervolt, and did A/B testing with Time Spy and BMI and in both I would get better thermals, slightly higher scores, higher average clock speeds, and a bit less wattage. I don't see a reason not to undervolt considering how easy it is to do and also revert in case you ever need to!

I wish the companies would give us a performance profile/bios and also an official undervolt profile/bios.
 
Last edited by a moderator:
Just posting results here tried 900mv at2025 crashed at 4 hours nlack ops cold war so lowered to 2000 and stays constant no crashes and 59c. This is on an aorus master 3080. Test xplane and msfs as well no crashing and59 to 61c on any game might try and up later to 2050 at 925
 
I love two TUF 3080s I put in my kids new rigs. Currently running lazy 1900@875mv (using MSIAfterburner 463Beta4Build1591). The Port Royal Score was ~11500.
 
What happens if you uninstall MSI Afterburner? Will your card go back to stock settings? If so, is there a way to automate this or modify the BIOS with those numbers or something that won't require relying on one piece of software?
 
mnewxcv, thanks for this guide. I’ve been trying to play around with my 3080 FE since I got it last week. I love the looks of it over the AIB models, which is one of the main reasons why I got it. However the noise is not the best as it gets rather loud in my case. So I’m trying to get it to a point I’m happy with but I’m not quite there yet. I’ve gotten better results with your method than others such as the flat line. With others I’ve had to do quite a large undervolt and underclock to get decent results, while your method has a better trade off imo. I know this thread is a little bit old but I had some questions.

1) when it comes to setting the power limit, I see that you don’t sacrifice performance very much with your method. However, I’m wondering what kind of pitfalls I may encounter with an 80% PL in games?

2) it seems that my GPU just loves to seek higher temps regardless of setting. Running time spy at stock settings will get it up to 77. So far the lowest I’ve been able to achieve is is a Tmax of 73, w +75 MHz and 85% power limit. Any tips on bringing these temps down further? I’m trying not to play around with case fans because I want to try to isolate the effect on the gpu first but I’ll certainly do that next.

Thanks for any help everyone!
 
mnewxcv, thanks for this guide. I’ve been trying to play around with my 3080 FE since I got it last week. I love the looks of it over the AIB models, which is one of the main reasons why I got it. However the noise is not the best as it gets rather loud in my case. So I’m trying to get it to a point I’m happy with but I’m not quite there yet. I’ve gotten better results with your method than others such as the flat line. With others I’ve had to do quite a large undervolt and underclock to get decent results, while your method has a better trade off imo. I know this thread is a little bit old but I had some questions.

1) when it comes to setting the power limit, I see that you don’t sacrifice performance very much with your method. However, I’m wondering what kind of pitfalls I may encounter with an 80% PL in games?

2) it seems that my GPU just loves to seek higher temps regardless of setting. Running time spy at stock settings will get it up to 77. So far the lowest I’ve been able to achieve is is a Tmax of 73, w +75 MHz and 85% power limit. Any tips on bringing these temps down further? I’m trying not to play around with case fans because I want to try to isolate the effect on the gpu first but I’ll certainly do that next.

Thanks for any help everyone!
Chances are you will not see major issues with 80% PL at all. The card will keep you at the appropriate clock speeds to maintain that. At that PL, you will likely see a mostly linear drop in performance of a few percent. Going below 80% may cause certain tasks or games to drop off moreso and cause issues like stuttering in some cases.

You mention complaints about noise and heat. Unfortunately they go hand in hand. If you feed the card cool air, by making your case airflow as efficient as possible, you should see a reduction in both heat and therefore noise. What case are you running?
 
Re: temps. I bought an AIO for $110. It crushes the heat, while being much quieter. Because I also have an AIO on my CPU, I had to get creative with the GPU radiator in order to make sure the it's higher than the pump. I'm not particularly concerned with appearance, and got it done. With an AIO, you can configure the GPU anyway you want and not worry about temps or excessive noise.
 
Chances are you will not see major issues with 80% PL at all. The card will keep you at the appropriate clock speeds to maintain that. At that PL, you will likely see a mostly linear drop in performance of a few percent. Going below 80% may cause certain tasks or games to drop off moreso and cause issues like stuttering in some cases.

You mention complaints about noise and heat. Unfortunately they go hand in hand. If you feed the card cool air, by making your case airflow as efficient as possible, you should see a reduction in both heat and therefore noise. What case are you running?
Thanks for the reply. My case is a Meshify C. I have 2x140mm scythe kaze flex as intake and 1x120mm noctua nf-s12 as exhaust. Toying around with another 120 mm right above GPU exhaust but doesn't seem to make that much of a difference.
 
Thanks for the reply. My case is a Meshify C. I have 2x140mm scythe kaze flex as intake and 1x120mm noctua nf-s12 as exhaust. Toying around with another 120 mm right above GPU exhaust but doesn't seem to make that much of a difference.
you could try setting you front fans to run faster in the BIOS, if they aren't already running at 100%.
 
What I did for case fans was to tie them to the PCIEx16 sensor on the motherboard (within the bios). Don't know if all motherboards have that option, but it works pretty good. Might have to set up custom curves also. The sensor on my board varies between high 40's and low to mid 60's (Celsius) . But it mostly ramps up with vid card usage, so that works pretty well. It is slower to respond. Ramps up over a few minutes and ramps back down similarly.
 
Hmm, is it possible for the card to override Afterburner curve and set its own voltage?
I have it set at 875mV and in the past when I would check HWInfo64, the max voltage would be reported at 875mV during gaming sessions.
But I just checked HWInfo64 now randomly after a gaming session and it reported a max 9xx mV
I relaunched MSI Afterburner, double checked that it's set to apply the values at Windows startup (it was), shrugged, closed Afterburner, launched Control and ran around a couple minutes to put some load on the GPU, exited and checked HWInfo64 - 875mv was the max reported voltage.
I reset HWInfo64 and went back to using the desktop and some hours later checked again and hwinfo64 reported a max of 900mv
 
Last edited by a moderator:
If you close afterburner, it will retain the most recent settings if you reboot. The GPU won't change it on its own.

Editing this to mention that it that once you apply the profile, rebooting the system without again loading afterburner, the settings persist. You could uninstall afterburner.
 
Last edited:
Will undervolting reduce the actual wattage? I did it with the 3090FE, but the overwall power usage stayed the same (350w) and the temps didn't change much, and I Was a good chunk lower (.975ish)
 
undervolting if done right will reduce watts = power. Read this thread from start. This enables you to get same or more performance @ lower voltage and of course lower temperature and noise. And it reduce the possibility to reach power limit of your card.
 
Will undervolting reduce the actual wattage? I did it with the 3090FE, but the overwall power usage stayed the same (350w) and the temps didn't change much, and I Was a good chunk lower (.975ish)
if you use method in OP, you need to set power limit to reduce wattage. If you use the method of setting a fixed voltage like 875mv, you should see less power draw because higher boost tables are no longer available.
 
if you use method in OP, you need to set power limit to reduce wattage. If you use the method of setting a fixed voltage like 875mv, you should see less power draw because higher boost tables are no longer available.

I used the curve in Afterburner, so the higher boost was still available, just at lower voltage :) Used GPU-Z to monitor, the voltage was lower, power was still pegged at 350w though.
 
Back
Top