But it's not just the 3090's. 

bmgjet
You have to lol when the only person providing any useful info is getting called out for not owning that card.

If u want the facts here the are.
The cards running a up9511r controller. The PCB is mass produced so they are all set up with the same config resistors, Which is what sets the balance.
It cant software rebalance the plugs and slot power like the other brands that used digital controlers which just need a I2C command sent to change it.
It wouldnt be a issue if every die was the same but all GPU dies arnt equal. There is this thing called silicon lottery so some will use more power or less power for the same task.
The die is made up of many different parts and each of these parts is also subject to the silicon lottery.
Only EVGA really knows how each part is wired up to the VRM. If one of those parts is really high leakage that inputs going to hit its power limit before the others.
When that power limit is hit the card wont go any further.
 
If you want to see what each inputs power limit is you can use my bios editor.


And yes I do have a EVGA 3090.
There is nothing he can do himself to change it with out voiding warranty.
Really this would of never been a issue if the power limit arms race didnt start and the cards stayed at there intended release day power limits which was 420-450W for the FTW3 cards and 320-366W for the XC3.
EVGA knows this which is why the 500W bios will always be a beta bios since a large ammount of people wont be able to hit that total board power that it advertises.

I've ran your tool, and checked my bios. The max 8pin pl is showing as 135000mW hmm. 

neteng101

donnie123
But is this really resulting in less performance? I mean I've tested 3dmarks and been scoring as high and above as average, surely it would be then weaker then average?

There are plenty of 2x8 PCIe power cards to skew the 3DMark benchmarks in favor of any 3x8 PCIe power card.  But to answer your question, a power imbalance will cause throttling (Power limit) if any one of the rails supplying power tries to go over its limit.  So if one of the PCIe connectors tries to go beyond 150w and the card is not as the BIOS total limit (eg. 450W XOC BIOS) then it will still cause a Pwr limit state to be set resulting in the GPU trying to reduce GPU boost to limit power consumption.  In the original GPU-Z picture this would be #2 PCIe connector that can possibly trigger the Pwr limit even though the board power draw isn't maxed out.
 
The power imbalance is why a lot of people complain they can't hit the board power limit they believe their card should be capable of.

Is this accurate?  I've noticed something like this. I hope this isn't the case.

Feklar
Is this accurate?  I've noticed something like this. I hope this isn't the case.

Read bmgjet's explanation.  It seems to be the case for cards that utilize uP9511 controller.  Here's my 3080 XC3 Ultra, heavily undervolted so I don't hit the power limit a lot but the two load peaks are Timespy Extreme GT1/GT2 tests.  The green lines (PerfCap Pwr) was seen when 8-pin #2 hits 150W, even though the card's power limit is maxed out the board power draw doesn't hit 366W.  My #1 gets to ~130W max so its always #2 causing the Pwr limit to trigger before the card can hit its board power limit.
 
The other thing is that the PCIe slot draw can cause problems - if you look at the slot voltage there is a drop under load - in my case it goes from 12.2/12.3V to 11.9/12.0V but in other readings I've seen shared like from a 3090 FTW3, it can drop to 11.7V and such.

Feklar
Is this accurate?  I've noticed something like this. I hope this isn't the case.

Yes, of course it is accurate. That's why I said it isn't a good thing.

I doubt the issue happens because of uP9511 controller, although not the best one, and the 12 would be better, many cards have it and it doesn't have this problem and it should work perfectly.

How do you find what controller your board is using?
 

donnie123
I doubt the issue happens because of uP9511 controller, although not the best one, and the 12 would be better, many cards have it and it doesn't have this problem and it should work perfectly.

 
Only 3 brands use that analog controller. (Galax, Zotac on there low end 2 plug cards.) and EVGA. And all 3 of them have people reporting issues of being unable to hit board power limit when using a bios with higher power limit then stock.
The rest used Digital.
 
 

MDG73
How do you find what controller your board is using?

Remove the heat sink and read the part number off it.
Or you can look in the 3090 thread on OCN. First post has a table of all the info.

3080's FTW3 ultras have the same issue btw.

neteng101

Dabadger84
*eats popcorn whilst enjoying a 3090 that knows how to power balance properly* 
 

Its very likely the uP9511 - the controller is as dumb as bricks and relies on an analog feedback loop from the GPU to tell it what to do.  Everything is set once it leaves the factory and there's no way to change the behavior via firmware.  This is very likely why we've not seen any fixes to the FTW3's dying out there, since they can't reprogram the power delivery logic on the fly.
 
Every card using the uP9511 seems to have imperfect power balance - my 3080 XC3 Ultra draws 25-30W less on #1 while #2 PCIe reaches 150W easily...  very much like the picture posted here for the FTW3, but at least it doesn't have a 3rd input that's super weak.  And it makes sense for the FTW3 given what the decoded MCU code showed in programming for the FTW3, the 3rd PCIe power rail was initially meant to be 6-pin ie. 75W capability.  Seems like EVGA changed the connector last minute, but forgot to rebalance the power delivery.  And now they have this big problem with FTW3 cards with no easy answers.
 
EVGA can release all the BIOS updates they want but the card can't magically realign itself physically to rebalance power.  Would agree that if someone was already spending all that money for a 3090 they'd want a Kingpin type card.

More than likely this is the issue. Check Buildzoids video down here at 4:38, same conclusion with that dumb controller:
 
https://youtu.be/ZuimvlNraLM