Michapolys
Alright, here goes:

There are currently two issues affecting modern high power GPUs, one relatively minor and one relatively major:

1. The minor issue is that some PSUs have misconfigured controllers, resulting in voltage regulation going haywire under some circumstances. The protections kick in, resulting in black screen. Units like the Seasonic Prime and Silverstone Strider Titanium were known to be affected and they were reportedly fixed on newer batches. Nothing we can do about this, apart from asking for a replacement. What concerns us here is the second issue.

2. Modern PSUs operate at variable frequency. When they are lightly loaded they operate at just a few hundred hertz, while when they are heavily loaded this increases into a few dozen kilohertz. The problem lies in the transition from low to high frequency.
In order to sustain a reasonable transient response timeframe, the output filtering capacitors of the PSU, as well as the input filtering capacitors of the load (in this case the load is the GPU) also double as hold-up capacitors for the load, for when the operating frequency of the PSU switches from low to high.
Unfortunately the way GPU manufacturers and PSU manufacturers interpret the ATX specification when it comes to transient response requirements differs, so the combined capacitance on the GPU input filtering capacitors and PSU output filtering capacitors is usually not enough.
When that is the case and there is a major sudden load from the GPU, all the capacitors on the +12v circuit end up emptying, voltage drops significantly, current increases dramatically, and when you are lucky, the PSU protections kick in in time and save the day (black screen). When you are not lucky, the protections take too much time, do not kick-in in time, and the GPU is physically damaged. When that happens, if you are lucky, just fuses get fried, if you are less lucky, FETs on the VRM get fried too, and if you are really unlucky, the GPU die gets fried. The worse part is, that even though some PSUs do actually have enough output capacitance to handle the transition, this comes at a cost. The high capacitance capacitors on PSUs are usually the wet electrolytic ones, and those can take extremely variable frequencies and extremely variable current without problem, but they cannot handle both simultaneously without significant hit to their lifespan. When put at such harsh conditions, their expected life decreases tenfold! So a wet electrolytic expected to last 10-15 years, would only last 1 or 2 years under such conditions without going out of spec and damaging our load.
In order to mitigate this issue, you want to increase the hold-up capacitance on your circuit, and since doing that with every PSU out there is impossible, the only way to do it reliably is by increasing the input capacitance on the GPU itself. This will not only increase compatibility with the PSUs out there, but will also save the card from the physical damage the lack in capacitance would cause under the circumstances.
NVIDIA expects the PSUs out there to deliver their power about twice as fast as they actually do. So doubling the size of the 16v 270uf input capacitors to 16v 540uf or more (Apaq has 680uf parts that fit the bill) should fix the power delivery issues, assuming the card uses the standard power limit. For cards with higher power limit (whatever uses a third 8-pin PCIE connector, like the FTW3), this should be increased even further to accommodate the increased power limit, meaning triple the capacitor size at 16v 810uf (Apaq has 16v 820uf parts that meet this requirement) to be safe.

... that is pretty much it.

If anyone fron the engineering department sees this, let me know if it ends up helping.

ty_ger07
And then there are all the other failures (the majority) which have nothing to do with the PSU and are instead related to the fact that MVIDIA's power monitoring is averaging based, allows spikes which exceed the video card VRM's current capacity, and the VRM's hardware protection was purposely disabled.

That's pretty much it, and that's the real problem.

ty_ger07
When you jump up from 100 watts to higher load, if your average power limit is 450 watts, what is your maximum power spike? Lots more. There is absolutely no monitoring or limit of spikes. If you exceed the VRM's current handling capability, eventually they fry. The VRM's hardware OCP was purposely set way too high because it would trigger too often. So, they are helpless against this problem. If you are lucky, the VRM's OCP will trigger and you will get a 100% fan black screen, but since the OCP was purposely set too high, often it is too late and damage has already been done.

https://youtu.be/rRp75YdLaq4
https://youtu.be/iTpKXJk8cAc

canhusrefoglu
So today I got my replacement card. Just installed it and works ok for now. Though gotta say, the initial momentary red light at the startup almost gave me a heart attack. Only issue is, despite my note, EVGA sent me another rev 0.1 card. So basically another ticking time bomb in my rig.
 
Starting on Monday I'll try my luck at Denver Micro Center. Wish me luck! 
 
And once again, I'd like to thank @rjohnson11 (not sure how to actually tag someone) for noticing my comments and keeping the ball rolling faster. 

rjbarker
^^^ Yep what mine did..."pop n poof" Red Light District....removed Water Block, re-install Air Cooler....RMA went smoothly..and quickly.
 
What kinda earks me is we pay around $2K CDN for a brand new Card...thing last 3 months...then I get a replacement "refurb" that I know nothing about...no history..what was the repair on this Card now in my $6k system....obviously EVGA's "warranty" includes replacing a defective product with a "repaired defective product"....would be a nice touch if there was some sort of "repair history card" included with the refurb they send us.
Refurb is working as it should...when I took the Air Cooler off to install the Vector Block I could tell the Card had never been taken apart before (based upon the hardness of the thermal putty n grease...like hardened ceramic), so who knows, maybe someone simply bricked the vbios on it....
 
The "pop" I heard was likely a pcb fuse within the power delivery....there was absolutely no visible damage on the pcb.....