my 3080Ti FTW3 Ultra with OC, ran very stable with my 5059x running stock clocks on a EVGA 850W T2

Just a little food for thought. The RTX 3090 can see peaks, depending on usage, that are 2~2.5 times the avarage current draw, and easily exceed the 600W~650W mark. These are certainly short term (milliseconds) and many well built PSU's can handle some of these. But current RTX models are certaimly on a different level than what we have been used to for decades now. Having said that, we do task our PC's with more than just GPU's and CPU's, we have ram, drives, AIO's, at times multiple case fans, peripherals on multiple USB ports... the totals can add up pretty quick and not leave any breathing room for headroom.

When I upgraded my RTX 3060 to a 3080 and 12th gen Intel, my 750W G2 was replaced with a 1300W. Using the 80% rule, where total wattage requirements at 100%, including GPU peaks, not just high averages and the rest of the items using the PSU came to just under 1000W, the 1300W at 80% is 1040W. But I do use this machine as a gamer rig, and the titles I run do fully stress both CPU and GPU.
 
So what does it look like when the PSU can't supply the peaks in a given game? Generally, the GPU BIOS will limit frame rate for the given frames it needs the power for, so it manifests itself in frame freezes or studdering. So if someone is seeing short term studdering, and you ask yourself, I've got a 3090, why is this happening? This could be a indication you might want to look at PSU capacity. Certainly if you do ever expereince some weird shutdown and can't explain why, look to the PSU for sizing.

Get an ATX 3.0 PSU and you will be fine. 850 to 1000 watts should be fine, with some headroom too.

DeltaBravo

Cool GTX

if CPU is under a normal - "high demand software load" = 750W -765W total for this Rig, as reported by my UPS, (the load is not one constant number (keeps moving) as the software's work load keeps changing

Steve over at GamersNexus did some testing on GPU load using amp clamps on the power leads with an oscilloscope and found that while the software denoted averages at full load, the software didn't report the actual demands during spikes, which were well over 2~2.5 times that. The EVGA RTX 3090 pulled ~400W and the TI model was 500W+ and TI OC was over 530W+ and that was just average not counting for intermittent spikes.

More so NVIDIA's design only monitors average power, therefore software can never expect to do better than that.

Intel's GPU design monitors and limits spikes and duration of spikes, like they did for their CPUs. Really a grand idea. Rather than ATX 3.0 and ATX 3.0+ and ATX 3.0+++ in 5 years, it would be great if NVIDIA cared to start monitoring and limiting power spikes. Their GPU VRM longevity would skyrocket mysteriously. It's been a reliability issue which appears to grow from one generation to the next. NVIDIA designs for average, adds some X factor of overhead, hopes for the best, AIBs push it, and then there are failures.

When it comes to power measurement, you have to use industry best practices to decide how you are going to measure and define your measurements. With a high enough sampling frequency over a short enough period of time, you can get power measurement nearly infinitely high. You have to decide what is a power spike that is long enough that it matters to the supply or load. In other words, you can kind of get any number you want, depending on how you measure it. I haven't watched Gamers Nexus video, so that number lacks context for me.