how much does case fan speed really impact thermals? 4x 1200-1500rpm a14's are going to be the major contributor of the overall noise profile. if you were to re-run e.g. the stock settings with fixed gpu and cpu fan speeds and step through the case fan speeds from 20%->25%->30% etc., when would you hit marginal returns?
Diminishing returns start around 1000 RPM for the NF A14s, and really, above 1250 rpm you don't get much gain at all. Almost everyone could just lock the fans at 1000 RPM, and run them silently, and unless you're *really* pushing a 3090's OC to the absolute limit, it wouldn't matter too much.
With OC'ing, a good part of the noise is coming from the A14's at 100%. However, In the particular OC that I have set up for the 3090, the fans have a significant effect on the heat dissipated via the passive part of the cooler in the middle of the card, which specifically affects GDDR6X memory temperatures. Going from 1100rpm >> 1300 RPM drops the 3090 fans from 1700 rpm to 1300 rpm.
High GDDR6X temps cause the 3090 to spin its fans up beyond 1700rpm, and the fans make a particularly annoying whine at those speeds. This is only an issue for memory intensive workloads such as ML training, but the noise is distinct and annoying enough that I wanted to avoid that. I could turn down my memory clock a bit, but with my core at 2Ghz, you really need the memory clocked higher to feed such a wide GPU.
To their credit, Nvidia balanced the card well. Overclock by 5% and you need to match that with a memory OC, or your performance gains will plateau. But you really do need to cool the memory.
------------------
Edit: Re-Ran OC Thermal Test with revised Fan Curve: 20% at 20C --> 80% at 90C. (Case fans at 1200 rpm)
* 161W on the CPU at 90C (vs 170)
* 69C on the GPU (OC'd as described before).
* Noise was 48dBA.
For a 5dBA reduction (53 --> 48dBA) , it was worth the 1.1% drop in CPU speed (3.81 vs 3.85 GHz all core), and 3% increase in GPU temps.
