Looking for 3090 undervolters!

[BetteBalterZen]

Hello!

Any 3090 users here who undervolts care to share their undervolting results? I'd like to compare with my results. 

My 3090 is stable at 1815MHz with 850mV and the result is good - very, very quiet(low RPM) while gaming and after many hours my GPU will occasionally hit 67c but usually it sits at 65c. My card draws about 30-ish watt less than with stock settings. I mainly wanna undervolt because I want my PC to be quiet with good temps but also because the 3090 is very power hungry. 

But after seeing Optimum Tech being able to hit 1800MHz with 806mV and 1850MHz with 850mV, I'm wondering if anyone else is able to hit those undervolts aswell? My 3090 can't do that, and it's a shame, because Optimum Tech reports his 3090 draws 71 watt less at 1800MHz using 806mV and that's pretty good. 

Picture is from Optimum Techs video on YouTube

[boggy77]

have you tried to undervolt to 800mV? if yes, what stable frequency are you getting for that voltage?

silicon varies, so it might be possible optimum tech simply had more luck with their card. or they got sent a better binned card by nvidia for reviews. overclocks and undervolts are rarely reproducible across the board.

 

[auraofjason]

Silicon varies but so does the program you're stress testing the undervolt with. For example in something old like heaven (what optimum tech tested with) I could do 1920mhz on 0.85v and be "stable" all day.  But in a game that maxes out the rtx cores like metro exodus I had to dial it all the way down to 1800mhz on 0.825v.

[BetteBalterZen]

23 minutes ago, boggy77 said:

have you tried to undervolt to 800mV? if yes, what stable frequency are you getting for that voltage?

silicon varies, so it might be possible optimum tech simply had more luck with their card. or they got sent a better binned card by nvidia for reviews. overclocks and undervolts are rarely reproducible across the board.

 

Honestly, no, I haven't tried lower than 1791MHz because I don't want to lose to much performance. But still, 1791 wasn't even stable on ~820mV. So I just settled in with 850mV and 1815MHz.

And yes sure, Nvidia could absolutely have sent a binned card and they probably did. 

[BetteBalterZen]

13 minutes ago, auraofjason said:

Silicon varies but so does the program you're stress testing the undervolt with. For example in something old like heaven (what optimum tech tested with) I could do 1920mhz on 0.85v and be "stable" all day.  But in a game that maxes out the rtx cores like metro exodus I had to dial it all the way down to 1800mhz on 0.825v.

Yes for sure. I have been able to complete full benchmarks runs in Heaven but when I stress test using the Time Spy Stress Test, the test will fail because of too many frame drops. I have 99.5% frame rate stability with my current undervolt. 

I like how Optimums 3090 is a FE version like yours. Maybe all FE are binned. Nice undervolt you have

[Beckenz]

On 2/19/2021 at 4:45 AM, BetteBalterZen said:

Hello!

Any 3090 users here who undervolts care to share their undervolting results? I'd like to compare with my results. 

My 3090 is stable at 1815MHz with 850mV and the result is good - very, very quiet(low RPM) while gaming and after many hours my GPU will occasionally hit 67c but usually it sits at 65c. My card draws about 30-ish watt less than with stock settings. I mainly wanna undervolt because I want my PC to be quiet with good temps but also because the 3090 is very power hungry. 

But after seeing Optimum Tech being able to hit 1800MHz with 806mV and 1850MHz with 850mV, I'm wondering if anyone else is able to hit those undervolts aswell? My 3090 can't do that, and it's a shame, because Optimum Tech reports his 3090 draws 71 watt less at 1800MHz using 806mV and that's pretty good. 

Picture is from Optimum Techs video on YouTube

What demo benchmark are you using to ensure stability? Do I need to test with max ray tracing on and off for stability?

 

I am getting my 3090 Strix (November 2020 batch) and I can post my undervolt for comparison.

[BetteBalterZen]

6 hours ago, Beckenz said:

What demo benchmark are you using to ensure stability? Do I need to test with max ray tracing on and off for stability?

 

I am getting my 3090 Strix (November 2020 batch) and I can post my undervolt for comparison.

I'm using Heaven Benchmark, the free version. I'll keep it running in the background(windowed) doing loops with the highest possible graphics settings + the highest possible custom resolution. Then I adjust the voltage curve / undervolt until Heaven freezes or crashes. When Heaven freezes or crashes, I bump up the voltage a tiny bit, until Heaven doesn't freeze or crash. Then I proceed to Time Spy Stress Test, which I have bought on Steam. If the Time Spy Stress Test have a framerate stability of 97.5% and up, it will complete the stress test successfully. If the test drops below the 97.5% in framerate stability, the test will stop (fail). I do, however, want to hit +99% in framerate stability. Is it important over the 97.5%? I don't know, it just feels good to have +99%  

So yeah; Heaven until freeze or crash and then Time Spy Stress Test

[Beckenz]

So I got my 3090 Strix and so far tested exclusively at 0.850 V using Maxes out RT cores demo. I am able to run this at 1920 MHz Clock, +250MHz Mem (Haven't tried to max this out yet), but at 430W power draw, and 70 C.

 

When I run those same settings on Heaven, I hit 380W, 62 C.

 

Will report back after I game a bit more and test with Time Spy Stress Test for framerate stability. 

 

Edit: Only got a 98.7% framerate stability. Which is "good" but I might need to tweak it to be better since I primarily game in VR.

[BetteBalterZen]

On 2/27/2021 at 3:47 AM, Beckenz said:

So I got my 3090 Strix and so far tested exclusively at 0.850 V using Maxes out RT cores demo. I am able to run this at 1920 MHz Clock, +250MHz Mem (Haven't tried to max this out yet), but at 430W power draw, and 70 C.

 

When I run those same settings on Heaven, I hit 380W, 62 C.

 

Will report back after I game a bit more and test with Time Spy Stress Test for framerate stability. 

 

Edit: Only got a 98.7% framerate stability. Which is "good" but I might need to tweak it to be better since I primarily game in VR.

1920 @850mV sounds really good man! Happy for you

[Beckenz]

On 3/1/2021 at 1:34 AM, BetteBalterZen said:

1920 @850mV sounds really good man! Happy for you

False positive :(

 

After running the Port Royal Stress Test again, I need to lower to 1830 @ 850 mV so I guess I'm not too far off from your results.

 

I also noticed that by default the card sets the target temperature internally to be 70C despite a temperature threshold being set in MSI Afterburner. So no matter what, it downclocks to meet the 70C. So I can run the card at 2010MHz @ 0.900 mV but within 20 minutes of a stress test it downclocks to 1965MHz.

 

Do you have the same issue of the internal temperature cap?

[Soap3205]

Im a little confused why you would buy a 3090 and then not use the extra performance over the 3080? 

[Beckenz]

15 hours ago, Soap3205 said:

Im a little confused why you would buy a 3090 and then not use the extra performance over the 3080? 

Even with the undervolt you still get a performance boost over the 3080. In order for a 3080 to match the performance of a 3090, the 3080 needs a greater power draw and temperature all things being equal.

 

For our usage cases, we are trying to find a preset where power draw is significantly less as well as the noise that comes from the GPU's fans. Obviously, we can use a more aggressive preset for more performance when we want to.

[BetteBalterZen]

On 3/2/2021 at 11:54 PM, Beckenz said:

False positive

 

After running the Port Royal Stress Test again, I need to lower to 1830 @ 850 mV so I guess I'm not too far off from your results.

 

I also noticed that by default the card sets the target temperature internally to be 70C despite a temperature threshold being set in MSI Afterburner. So no matter what, it downclocks to meet the 70C. So I can run the card at 2010MHz @ 0.900 mV but within 20 minutes of a stress test it downclocks to 1965MHz.

 

Do you have the same issue of the internal temperature cap?

Oh okay, sorry to hear.

 

I haven't tried overclocking the card, I just want good quiet performance
Only tried maxing out the power slider. Card ate 480w in an instant lol. 

[BetteBalterZen]

On 3/3/2021 at 3:08 AM, Soap3205 said:

Im a little confused why you would buy a 3090 and then not use the extra performance over the 3080? 

You say that like I'm making my performance worse than stock. My 3090 performs better than stock now.
 

[BetteBalterZen]

On 2/24/2021 at 5:37 AM, Beckenz said:

What demo benchmark are you using to ensure stability? Do I need to test with max ray tracing on and off for stability?

Hello again!

I've found this awesome - and free - Ray Tracing benchmark, made by MARvizer 

Link: Pure RayTracing benchmark by MARvizer (itch.io)

It totally maxes out the GPU and RT cores. Actually so much, that I had to increase the powerslider in Afterburner, to keep the 1815MHz from my undervolt. 
Under the test, with my previous undervolt setting, that used 100% (stock) on the powerslider, my GPU clock dropped from the 1815MHz to 1750-ish MHz, because maximum power was hit (the 100%). So I increased it to the 137% in Afterburner and that kept GPU clock on 1815MHz but total board power increased from 334 watts to 380-ish  watts. 

My card is still totally stable, with the undervolt and maxing out the Ray Tracing in the benchmark.

I used 2560x1440p and Pro(highest) settings, and got an average of 35.28 FPS.

Thought I'd head back to this thread and mention this awesome test/benchmark  

[tinsukE]

Just recently bumped into this thread and I totally agree that Optimum Tech's results seem to be good only for light rasterization workloads.

 

I had a 3080 TUF that, according to my old notes, was stable at 1800MHz and 837mV.

 

Now I own a 3090 TUF OC that is able to hold 1800MHz at 812mV, quite a nice result if I may say so:

 

1800@812 -> Super fine!

• Time Spy ST: 98.8% 330-340W Stable frequency
• Port Royal ST: 99.6% 340-350W Stable frequency
• Pure RT: 34.5 FPS 360-370W Stable-ish frequency (some drops to 1785MHz)

I've tried stretching it to 1860MHz, but it starts warming up more and becoming not so stable (frequency wise), I also am very concerned by any temperature increase since the VRAM chips on those cards run very hot:

 

1860@850 -> Fine-ish

• Time Spy ST: 99.4% 350-370W Stable frequency
• Pure RT: 34.3 FPS 370-375W Unstable frequency (1740-1815MHz)

Nice tip about this Pure RT benching tool. I think that anything that passes Port Royal's stress test can pass it too, but hey, it's free!

 

@BetteBalterZen how did you get that Pure RT result of 35.28 FPS? Mine is a tad lower but I left the GPU fans on auto, with the latest Resizable BAR BIOS, and I also have an air-cooled 5950X on stock.

[BetteBalterZen]

On 4/6/2021 at 10:34 PM, tinsukE said:

Nice tip about this Pure RT benching tool. I think that anything that passes Port Royal's stress test can pass it too, but hey, it's free!

 

@BetteBalterZen how did you get that Pure RT result of 35.28 FPS? Mine is a tad lower but I left the GPU fans on auto, with the latest Resizable BAR BIOS, and I also have an air-cooled 5950X on stock.

You're welcome  

And I was able to hit 35.28 by maxing out the powerslider while using my undervolt. So my settings was 1815MHz @850mV, with 137% powerlimit. 
By increasing the powerslider to 137%, my 3090 was able to stay at 1815MHz at all times. This is probably why my score is higher than yours. 
Also, if your undervolt settings is just enough unstable to drop a frame or two in total under the full run, that could lower the score aswell of course. 
My system at the time of benchmark: 8700K @4.3GHz, 32GB 3600MHz CL16 RAM, no Resizeable BAR (my motherboard does not support it). 

See my photo for benchmark result: 

[tinsukE]

Aha that might be it, then. Afterburner only lets me increase the power limit to 107%, I guess that's one benefit of having a STRIX over a TUF =]

[BetteBalterZen]

Just now, tinsukE said:

Aha that might be it, then. Afterburner only lets me increase the power limit to 107%, I guess that's one benefit of having a STRIX over a TUF =]

I think so yeah

[kaiwenwu]

kp bios let me run over 2000 at 0.7v

[BetteBalterZen]

9 hours ago, kaiwenwu said:

kp bios let me run over 2000 at 0.7v

Wut, is that stable??? That can't be real. 

[kaiwenwu]

11 hours ago, BetteBalterZen said:

Wut, is that stable??? That can't be real. 

The problem is the stock vbios don't let you go that high, I mining at 2000 is fine. The hash rate is about 1m higher. 

[kaiwenwu]

11 hours ago, BetteBalterZen said:

Wut, is that stable??? That can't be real. 

If I go below 1800 the hash rate will drop a lot so it is kind strange. 

[BetteBalterZen]

7 minutes ago, kaiwenwu said:

The problem is the stock vbios don't let you go that high, I mining at 2000 is fine. The hash rate is about 1m higher. 

Aha, so you haven't tested 2000MHz at 0.7mV in games or stress tests? 

[tinsukE]

Believe it or not, today I got my hands on a 3090 FTW3 Ultra and I gotta say I'm disappointed.

 

I expected the card to be power hungry at stock, and it was, by sucking up to 400W to achieve 1900MHz on Heaven.

 

I rolled up the sleeves and got down to tweaking, it seems pretty stable at 1800MHz with 800mV. But the temperature is just comparable with my TUF OC (up to 75C), consuming around the same amount of power as the TUF (up to 320W), but much noisier at stock fan speeds. Tweaking the curve a bit brings it to 82-83C at a slower fan curve that I find comparable noise wise to the TUF's stock one.

 

It seems that the power hungry design does make it run hotter. Some also say that the cooler also keeps the VRAM cooler (which causes the GPU to run a tad hotter), but I see the same overall 94C for the VRAM too.

 

I shouldn't have opened its box. I was happy with the TUF and didn't know it. I guess I'll re-sell it with a smaller profit than it could net me sealed.

Edited April 9, 2021 by tinsukE
Better formatting