[SOLVED] MSI 3090 Suprim X undervolting questions

[Adam Jelinek]

Hi guys so the 3090 is a very hot running card and it definitely makes my room quite toasty. I’ve been reading ever since the launch of these cards with instability and such issues about undervolting the cards. I’m not quite sure how to do this, or even what numbers I should undervolt to. Mainly I would like to reduce the heat coming from the card, as well as maybe reduce the temps a little bit. Thank you for the help as well.

 

[Phaaze88]

There's not a lot you're going to be able to do about this up to 450w monster without spending chunks more money, or taking a performance hit.

Considering how MASSIVE Msi made the cooler, it isn't that good compared to other models.
It may need to be changed if it's the weak link, and the only alternative there is a custom loop... [That's where the 'chunks more money' comes in, 'cause custom loop isn't cheap, nor is it a simple step1, 2, 3; it requires planning in advance.]

Still have that O11 Dynamic?
-Bottom + side intake and top exhaust?
-Bottom intake and side + top exhaust?
^Might have to experiment and see if the one you're not doing helps temperatures at all.

Trying a different paste MIGHT help. If you're not comfortable with the idea, that's fine. Skip it.
The thicker pastes work better on the gpu die. Because of its glass-like surface, they resist pump out from cooler mounting pressure better.
Some samples include: Noctua NT-H2, Cooler Master Mastergel Maker Nano, Thermal Grizzly Hyrdonaut, and Gelid Solutions GC-Extreme.
I do not recommend Thermal Grizzly Kryonaut unless the gpu is liquid cooled. This stuff 'burns out' faster, the higher the temperatures are - around 80C - leading to more frequent repastes.


I'm not even going to touch on memory pads, because I've seen/read mixed results from that.
It's all over the place, because there's no set standard testing conditions between different models of pads.


Undervolting may help, or may not help - it's still an up to 450w card - it's also TEDIOUS to do. It's going to be easier to just open Afterburner and run with a lower power limit... but that also means lower performance.
I've been playing around with it for a few weeks on my 1080Ti. Here's the 'short version'.
-Msi Afterburner and Gpu-Z(for the Sensors tab) are needed.
-Afterburner has save slots for profiles. Make use of them.
-With Afterburner and Gpu-Z open and running in the background, play your games. Simple as that. The goal is to find out how much voltage(max) your gpu is asking for.
That will be your 'starting point'. Whatever that number is, write it down or memorize it.

-Go to Afterburner's Curve Editor. A window will pop up containing the Voltage/Frequency curve for the gpu. You will see several points on said curve. This is all the voltage control Nvidia has allowed for.
-You can left click any point and lock the voltage there with the L key, as well as raise/lower the frequency from that point with the Up and Down keys.
-Recall the max voltage requested by the gpu, and find it on the chart, then select and click on the point to the left of it. [You can go to points further left if you wish.]

You'll notice as you go further left, the boost frequency gets lower. That's not the worst of it though.

I could just end it right here - it would still keep things simple: Curve Editor, experiment with points lower than the max requested, lock it, click Apply on the Main HUD, and go game with a little less performance...
But you probably want to offset some of the lowered boost frequency so as to reduce some of the performance hit, and this is where it becomes a pain in the butt - and Gpu Boost doesn't help, because it 'whines' over every little thing:
-Every 5C or so warmer that the gpu runs(excluding thermal limit)? "Imma throttle the boost frequency back a little."
-Don't have enough voltage to run at a particular frequency? "Imma throttle the boost frequency back a little."
-Keep bumping into the power limit? "Yeah..."


-If you're going to increase the frequency, do in increments of 10-15mhz.
-Test the frequency bump for stability across as many games/benchmarks as you are willing. The more samples you use, the better.
You can use Unigine Heaven, but honestly, it is out of date - it doesn't even have DX12 support.
You can run Unigine Superposition, but it is very limited with the free version. If you don't want to spend extra, the first suggestion for testing is the best.
EXCLUDE Furmark and Kombustor. They do not run like actual games, and are better suited for testing the capability of the gpu cooler - they're like Prime95(small FFT, all AVX off), but for gpus.

-You basically keep raising the frequency - if you even get that far - until stuff starts crashing, and you dial it back to the last stable increase.

~And, I think that's it?
Unfortunately, I don't believe any set numbers exist, because of silicon lottery and Gpu Boost.
Memory clock? Nah, leave that one alone. Frequency is all the control we have on this one. Lowering it does not help temperatures: the voltage is fixed, and it doesn't pull much power to begin with.

Crap. I got carried away again...

 

[Juffbox]

You can try starting at 1860-1875 Mhz and 850 mV and then go from there, from my own OC and UV experiences my card (3090 strix) wasn't too far off from stable at that point.

Also, make sure to use a very demanding game and not just benchmarks to test. I found cyberpunk to be a good game as it uses both DLSS and ray tracing, so everything gets tested. If you can go 2 hrs without any errors it's probably stable in other games as well.

 

[Rukeith]

In MSI Afterburner's curve editor, take whatever your stock clock is and bump up different voltages starting from 1.0v going down to 975>950>925>900 etc until you find a voltage that's stable while something like heaven benchmark runs in the background. That way you can find the lowest voltage your card can handle while also not losing much if any perceivable performance.