What i was saying, it that it takes certain load and temperature, for an air cooler to reach peak efficiency. And it doesn't happen at 40-55°C (assuming 20-25°C ambient), which is what you get on liquid cooling. For the GPU scenario, it usually happens around 55-70°C.
I'm aware that heatpipes/vapor chambers have a tighter operating temperature range than a water loop, and are generally more efficient near the higher end of their range, but I'm doubtful this is a major factor in most GPU cooling situations. If the main limiting factor is heatsink fin area, the deltas aren't going to be much different along the practical heatpipe operating range.
On this Gigabyte 3080 with the modified stock cooler (which has about three times the fin area of the FE cooler) using new TIM and three Arctic F9s at ~1500rpm, full load (370w limit) temperature deltas are around 30C ambient to GPU edge, 42C GPU hotspot, and ~50C for the GDDR6X junction.
I'm pretty sure it's not under noise normalised conditions, cus otherwise your custom loop build is poorly configured, which is not uncommon or unheard of. It's actually quite common lately, since 2x360 or similar capacity, it's not oversized anymore for GPUs that go between 300 and 450watts and CPUs with the behaviour and thermal density of ryzens... So the faults in user configuration, and choice of case, fans and cooling parts show up quite easily.
It's fairly well noise normalized, subjectively speaking (I didn't do any measurments, I'm just running the fans up to the level I start to be able to distinguish them from background noise). However, I am running rather light on rad area for a 3080; a single 240x30 14 FPI rad with pair of Arctic P12s (normally running at 1200-1300 rpm, mounted as a front intake, with a DDC pump at 40%/2000 rpm (which is complete overkill, but I'm planning on adding a CPU block and 280 or 360 radiator to the loop, once I get things in a more permanent configuration). There are no other components currently being cooled on this loop.
At the same load and 370w power limit, the temperature delta advantage for the loop is maybe ~3C core/hot spot and ~8C GDDR6X (which is a bigger advantage than it seems because the memory on this FE clocks higher than any other non-hardmodded 3080 I've ever seen...+1750MHz or so, stable).
Flowing water can potentially radically outperform heatpipes in heat carrying capacity and it's much easier to add radiator area to a loop than fin area to a heatpipe cooler, but if vaporchamber/heatpipe capacity isn't the limiting factor and fin/rad area (and air flow through it) are similar, then temps should be roughly similar...and in my situation, they are.
Sure, I could triple the radiator area of this loop and knock another 10C off temp deltas with dead silent fans, but there is little incentive for me to do so. The main reason this FE is even watercooled in the first place is because the stock cooler and TIM combination made the card almost unusable due to memory temps maxing out and the FE cooler is obnoxious enough to work with that I was disinclined to limit myself to a pad swap.
With the sort of power limits I'm working with, a big air cooler with real fans on it is plenty sufficient. Stock heatsinks are pretty much there on high-end GPUs, but they still have heavily integrated slim fans that tend to be loud for their air flow and a chore to replace.
What's important is that the heatsink is wide enough to support 120mm fans. I lost a few degrees rigging a12x25s on my card, so it isn't a guaranteed drop in temps.
Ducting helps a lot. I have A12-x25s on my PowerColor 6800 XT and there was a ~7C drop in temps when I rigged up a posterboard shroud to make sure all the air coming out of the fans was passing through the heatsink fins.
Of course, the stock fans on many coolers can ramp up to 3000-4000 rpm and it will be hard to beat full speed temps, but almost no stock GPU fans are going to move as much air as a pair of A12x25s at anywhere near the same noise level.
.
