Hey all, I recently received my EVGA RTX 3090 XC3 Ultra and I have it installed in my NCASE M1 v6 now and I am interested in improving the temperatures and noise profile by undervolting. I am a bit confused at my results so far and not sure if I am going about this in the correct way. I am using MSI Afterburner to adjust voltages. The Voltage/Frequency curve editor shows a max 1950 MHz frequency at 1100 mV voltage in the curve at stock (see attached screen cap) but as I understand it the max boost clock on this card is 1725 MHz? When watching the Optimum Tech how-to he says to drag down the entire curve to the rated boost clock and then ramp up the lower voltages in the curve to this value until I hit a minimum voltage that a stress test like Heaven will run on. Seems like when I do that I am still getting GPU temps near max (mid-80's C) and lower max clocks. Fan noise and thermals are not where I would expect them. Any help is appreciated!

Thanks for the reply! Is it too much to say I’m bothered by the fan noise whenever it exceeds 50%? When I set my fan curve as you say so that fan speed does not exceed 50% once past 50C and offset and under volt as you describe then run Heaven the temps climb well into the high 80’s at which point I begin getting concerned and stop the test.

The fan speed being annoying depends entirely on the tone & loudness of the fans on the card.  My Kingpin in particular has one of the more annoying "tones" I've ever heard when the fan is at high speed, it's nowhere near as loud as an old blower-style card in volume, but the tone of it is far more annoying than those things ever were, at least to me, at maximum speed.
From what I've heard on other people's Kingpin reviews like Luumi, the Kingpin on-board VRM fan can vary in noise tone/level quite a bit, and it seems mine is one of the "bad" ones lol - The only way I could describe mine is... like one of those old school chipset fans, that higher pitched WEEEEEEEEEEEEE they made... only it's much louder in volume.  Luckily it only happens above 80%, so I just set my fan curve for the on-board VRM fan to top out at ~75% unless the GPU core goes over 55C (which doesn't happen) - that's right on the brink before that loud tone kicks in, and it keeps the VRMs/components the card cools inside the shroud plenty cool.
Whereas Luumi's might be the quietest one I've heard: https://youtu.be/2QIoeeUW5oc?t=1106
However, 50% fan speed on an air-cooled card is very likely not gonna cut it, since the default fan curve, I believe, on most of the 3-fan cards takes it up to around 55-60% for a "maximum" speed. 
One way to test how much of the issue is raw airflow problems would be to test the card at a few fan speeds with the side of the case closest to the GPU on vs off, and see how much of a temperature difference there is.  Try it at 50%, 80% and 100% - give the card time to cool down between each test run, and run something that's got a decent length to it, like Unigine SuperPosition or Heaven or Valley - preferably SuperPosition as it's the most stressful of the 3.
If the temps don't go down noticeably between 50% & 80%, the case definitely has airflow issues, especially if the temps don't go down with the side of the case on, but do with the case off, between those two settings.
If the case has gaps/air ventilation on it, definitely setup any fans you add as intake, it's almost always better to have more intake vs more exhaust, more exhaust creates a vacuum which can cause a number of issues like making your card recirculate air or the like, whereas most of the time "too much intake" is not something that's actually bad thing.

Ok here are some results from additional testing to see to what extent the case/airflow is an issue. The tests below were run in the order listed at stock GPU settings using Unigine Heaven for stressing. In between each test I ran fans at 100% until GPU temps artificially stabilized below 30C, usually 26-29C. The left case panel is the one covering the GPU and right case panel is the one covering the back of the motherboard. Room ambient temperature is 23-24C. 
 
Test 1 - Left, right, and front case panels off

• GPU temperature @ 10 min - 79-80C
• GPU fans (auto) @ 65% ~ 2150 rpm

Test 2 - Left and front case panels off

• GPU temperature @ 10 min - 79-80C
• GPU fans (auto) @ 67% ~ 2210 rpm

Test 3 - Only left case panel off

• GPU temperature @ 10 min - 80-81C
• GPU fans (auto) @ 67% ~ 2240 rpm

Test 4 - All case panels off (left, right, front, and top)

• GPU temperature @ 10 min - 78C
• GPU fans (auto) at 63% ~ 2090 rpm

Test 5 - All case panels on

• GPU temperature @ 10 min - 84-85C
• GPU fans (auto) 76% ~ 2510 rpm

Obviously with all of the panels on it is creating a bit of a heatsink in the case over open of about 5-6C. After the above tests I decided to run one more test with all of the panels on and setting the GPU fan speed to 100%. Again, before the test I ran fans at 100% until GPU temps artificially stabilized below 30C, usually 26-29C.
 
Test 6 - All case panels on and GPU fans set to 100% ~ 3300 rpm

• GPU temperature @ 10 min -73-74C

I will also note that max clock recorded during all of these tests was 1920 MHz but that was a rare spike. Typical clocks during stress ranged from 1710-1740 MHz at stock voltage and 100% power limit/83C temperature limit.
 
Thoughts?
 
System specs:
CPU: Intel 10900k 
GPU: EVGA RTX3090 XC3 Ultra
Motherboard: MSI MEG Z490 Unify Mini-ITX
Memory: Crucial Ballistix 3200 MHz DDR4 DRAM 32GB (16GBx2)
CPU Cooler: NZXT Kraken X53 240mm AIO
PSU: Corsair SF750
SSD: Western Digital WD Black SN750 NVMe M.2 2280 2TB
Case: NCAS M1 v6.1

post approved

mlstoltzfus

Here is the default fan curve for the GPU.



The fan speeds of the curve and the ones you reported seeing in tests is about 10% lower than the curve...  what fan speeds are you taking in those tests, average?