I use an eGPU, so I have a 4x instead of a 16x PCIe 3.0 link with a 1080Ti and even that doesn't make much difference, some games blit/do more system <-> gpu transfers than others (APEX, I'm lookin at you) and there's a noticeable perf impact for the most part it's negligible. I wouldn't worry at all about PCIe 3.0 vs 4.0.

Same. In gaming and graphics benchmarks on my 1070Ti over Thunderbolt 3 (Hades Canyon Intel NUC), I've seen no more than a 10%-20% drop in frame rate vs. what I'd expect from equivalent benchmark results reported online, so I'd expect no material difference between PCIe 3.0 x16 and PCIe 4.0 x16 for typical gaming applications, and, from benchmark numbers I've seen, only a modest performance drop from PCIe 4.0 x16 to PCIe 3.0 x8.

This assumes, of course, that the PCIe lanes in question actually exist† and are dedicated to the card, and are not shared with another in-use device via a switch or similar chipset shenanigans.

Case in point: the only time I ever had performance problems with my eGPU setup was when attempting to do 4K HDMI output through a Blackmagic Decklink card while simultaneously using the 1070Ti for compute over the same (shared) Thunderbolt bus in DaVinci Resolve.

As far as I know, this would not typically be a problem with a single-GPU, non-Thunderbolt-connected desktop system, as desktop motherboards typically have at least one full-bandwidth x16 slot, which you'd almost always use for the GPU.

† Physical x16 slots can have fewer than sixteen connected PCIe lanes. For example, while my HP Z820 has four "x16" slots, only three have full bandwidth; the fourth slot supports x16 cards, but only at the speed of an x8 slot, because only eight of the slot's sixteen PCIe lanes are actually connected.

You can find the information from multiple sources - IT review sites that performed the testing - that for current GPUs PCIe 4 x16 is not needed, you can use PCIe 3 x16 with negligible impact and even PCIe x8 with minimal impact.

There are also some sites that explains why large capacity power suppliers are a waste of money. Ex: https://www.youtube.com/watch?v=X_wtoCBahhM

Great to hear that PCIe 4 is not a necessity at this point.

About the PSU, for me the only reason to go with a 1000W PSU is that the fan will remain quiet with a GTX 1080 when idle and gaming as well. At least thats what I hope.

If the same is possible with the 3080 is more questionable, especially if the card draws 320W at all times.

That capability exists even on smaller power supplier, for example the Corsair RM-650 has it; powering the fan is done based on temperature (and that depends a lot on the load), not on the power supply max rating. This means you need to just add your max CPU consumption (not TDP, that is lower), GPU max consumption, the rest of the components and add a 200W reserve, not a 500W reserve.

PSU efficiency is low at the ends of the curve, underutilized (<20%) the efficiency can drop as low as 70%. The top (not peak) efficiency is somewhere in the 30-80% load zone, that is where you want to be.

I was under the impression that it was the power draw alone that decided when the fan kicked in. But it makes sense that the temperature also matters.

Based on that info and that the system will draw about 430W for CPU+GPU, it seems that the fanless Seasonic TX-700 is better suited.

It is quite expensive, but as someone who really hate unnecessary noise, it looks like a good choice.

Changing all the existing cables to a new PSU sucks though :p

The major difference is offloading some of the load from the CPU when using the pci-e bus.

If you have a spare core not being used when playing a video game, it can be used which should make the difference between 3.0 and 4.0 somewhat indistinguishable.