My friends, they've ignored this for 2+ months on the FTW3. Time to stop being an apologist (hopefully wwxww can learn to read his own definition), and understand that EVGA (moderated for language) up hard this generation and hardcoded an analog ratio in the load balancer chips that they cannot change on these initial/1st generation PCBs.
 
For multiple generations now Nvidia PCB designs have included chips that dynamically route power from 8-pins + pci-e slot to VRM stages, to enable even loading of VRMs and uniform pull from pci-e psu cables. Some chips have had digital controllers that could be adjusted via memory scan or bios/firmware updates, but EVGA appears to be using analog chips that take hard-set input voltages on chip pins to control the ratio (bearded hardware manually hardmodded this voltage to adjust ratio as part of his FTW3 unlocking). Note that this is separate from the voltage controller, which is digital on the kingpin and analog on the FTW3. Now, each 8-pin and pci-e slot have an individual power limit, measured via shunt, and the core (nvvdd+msvdd) and vram (fbvdd)  each have an individual power limit, separate from the overall board power limit. The pci-e slot limit is around 80W. The hardcoded ratio on the balancer chips is set to low, that when the pci-slot is at 80W, the 8 pins are far below their individual power limits of 150W each. The reason Furmark etc can pull more is they pull way more power on VRAM, and the VRAM VRM stages do not appear to be balanced against pci-e slot like core does. 
 
On a solution note, it looks like either RMA until you get a batch where they've adjusted the ratio on the chip, or since we've got the standard 7 shunt assortment you can just shunt the pci slot 5MO resistor like FTW3 (the blue circled on in bottom image). It's a wider shunt to accommodate the well isolated traces and I'm working on figuring out the best resistor to buy from digikey etc to stack or replace. 

Pretty solid explanation you give here. Do you know why some cards seem to be affected while others are not? 

Chamidorix
My friends, they've ignored this for 2+ months on the FTW3. Time to stop being an apologist (hopefully wwxww can learn to read his own definition), and understand that EVGA up hard this generation and hardcoded an analog ratio in the load balancer chips that they cannot change on these initial/1st generation PCBs.
 
For multiple generations now Nvidia PCB designs have included chips that dynamically route power from 8-pins + pci-e slot to VRM stages, to enable even loading of VRMs and uniform pull from pci-e psu cables. Some chips have had digital controllers that could be adjusted via memory scan or bios/firmware updates, but EVGA appears to be using analog chips that take hard-set input voltages on chip pins to control the ratio (bearded hardware manually hardmodded this voltage to adjust ratio as part of his FTW3 unlocking). Now, each 8-pin and pci-e slot have an individual power limit, measured via shunt, and the core (nvvdd+msvdd) and vram (fbvdd)  each have an individual power limit, separate from the overall board power limit. The pci-e slot limit is around 80W. The hardcoded ratio on the balancer chips is set to low, that when the pci-slot is at 80W, the 8 pins are far below their individual power limits of 150W each. The reason Furmark etc can pull more is they pull way more power on VRAM, and the VRAM VRM stages do not appear to be balanced against pci-e slot like core does. 
 
On a solution note, it looks like either RMA until you get a batch where they've adjusted the ratio on the chip, or since we've got the standard 7 shunt assortment you can just shunt the pci slot 5MO resistor like FTW3 (the blue circled on in bottom image). It's a wider shunt to accommodate the well isolated traces and I'm working on figuring out the best resistor to buy from digikey etc to stack or replace. 

What is a safe average power draw from the pcie slot assuming I shunt it? My x299 board has additional power going to the pcie lanes coming from the psu, will this alleviate the dangers of excessive power draw from the pcie slot?

Chamidorix
My friends, they've ignored this for 2+ months on the FTW3. Time to stop being an apologist (hopefully wwxww can learn to read his own definition), and understand that EVGA ****ed up hard this generation and hardcoded an analog ratio in the load balancer chips that they cannot change on these initial/1st generation PCBs.

Imagine having to shunt-mod a Kingpin to make it worthwhile...
 
If true, I'll be returning mine. Sucks because I cancelled my other back-order that I was waiting 3 months for. Good god this GPU launch is enough to turn a man to another hobby... I mean really...!

I would not shunt-mod the card (especially don't shunt mod the PCIe power, that is actually a very bad idea itself), if it doesn't hit the power targets it should be hitting, RMA it - that's my opinion, as someone who's planning on buying one.

How exactly do you plan to just return it?

@Camidorix       It seems I misinterpreted your first post. Thanks for the informative post for the uninformed. i.e. me.
 
How would the card react to the shunt modification when not under load? Would the power draw be any different from the entire card with little load?

I am unfortunately having the same issue :(

Dabadger84
I would not shunt-mod the card (especially don't shunt mod the PCIe power, that is actually a very bad idea itself), if it doesn't hit the power targets it should be hitting, RMA it - that's my opinion, as someone who's planning on buying one.

^ this
If the product has a design issue, give EVGA the opportunity to fix it. this isn't an apologist POV, but more so a level headed consumer. 

Cadillac94pimpin
Pretty solid explanation you give here. Do you know why some cards seem to be affected while others are not? 

It is quite simple, they revised and fixed the problem on newer manufactured pcbs. As stated, the load balancing chips take an input voltage to determine the ratio so they simply adjusted this input voltage, or similar. It's already appearing more and more with newer FTW3s (an increasing number of reports that do 500W with the XOC bios), the very simple way to check is if you pci-e slot is drawing ~75+W or 50-60W under time spy extreme load. PCBs are manufactured in batches and there certainly are multiple batches of KPE PCBs that have been created, so it follows that you can reasonably assume the newer ones have the correct balance while the older ones made before the whole FTW3 XOC debacle (or whenever EVGA became aware) still have the problem. 
 
My KPE edition finally arrived today, so I will be testing the PCI-E draw and asking for an RMA if I have this issue. I'd recommend all other KPE owners do the same (run your card under TSE on 520W bios and observe PCI-E slot power draw in GPU-Z or HWInfo), to hopefully pressure EVGA into providing more support or clarifying if there is a misunderstanding somewhere.