3090FE GPU Clock stuck at 210 MHz Board Power Draw 390Watts at Idle. PerfCap Reason Pwr

[HiResJunky]

Hi all, really banging my head here hoping someone can help. I have scoured the interwebs and read countless threads on similar issues but none of the solutions are working for me.
Background: I'm upgrading a 5-6 year old custom water cooled system from 1080ti to 3090FE.

Specs: ASUS Maximus Formula X Bios 1801
Intel 8086 oc'd to 5.2MHz(stable for years)
32GB Ram @ 3300
ASUS PG348PQ monitor. G-Sync enabled 3440-1440 locked at @100mhz using Displayport
Seasonic Prime Titanium 850W(using 2x 6+2Pin PCI-e Connectors)
RTX 3090FE with EKWB Waterblock(just installed waterblock card not tested before installing)

Note in GPU-Z screen shot 8-pin #1 Pwr is low and 8-pin #2 power is 380+ watts at idle

Things I have tried. DDU uninstall GPU drivers in safe mode
Uninstalled ASUS Armoury Crate
Uninstalled reinstalled MSI Afterburner
Uninstalled reinstalled Latest Nvidia drivers

Questions Is my power supply the issue? The supplied Nvidia 12pin power adapter splits to 2 8pin pci-e connectors. I have plugged in 2 6+2 pci-e cables but noticed on the PSU side of the cable there is only 6 pins not 8. Is that normal?
I have also tried have a custom white GPU extension cable see pic, only 6 pins instead of 8 is this normal?
I have another spare newer Seasonic power supply Prime 650W and noticed the 6+2 pin PCI-e cables it comes with have 7 pins on the PSU side. I tired these in the 850W power supply, makes no difference.
Motherboard PCI slot @x8 3.0 not enough for 3090FE?

Thanks a bunch, if there is anything else I can include that can help let me know.

 

[oxrufiioxo]

Normally the card comes with a 3x 8pin to 12 pin adapter. For some reason yours isn't reporting the Ram power draw could be a bug I guess 400w for the core sounds about right memory would be in the 100w range. Can you run something like Timespy just to verify your are getting ballpark performance for a 3090?


edit that was the 3090ti..... I guess it was 2x 8 to 12pin but still seems the issue is the memory. The 400w at idle also sounds like a bug your temps would be way higher lol.

It definitely has nothing to do with the wattage of your PSU My 850 prime TX in my secondary pc handles a 450w gpu no problem.

Guessing the card was used when purchased did you try contacting the original owner if that was the case?

 

[pavle]

Looks like as if there is something wrong with the shunts for measuring current, if true, too soon after the card was new; definitely ask the original owner what was going on with the card.

Edit: Those missing pins on your 8 pin connectors should really be connected, they are ground pins also could be the reason for false reading, sometimes, if implemented, they are referred to as sense pins (at least 1 of them).

 

[HiResJunky]

Normally the card comes with a 3x 8pin to 12 pin adapter. For some reason yours isn't reporting the Ram power draw could be a bug I guess 400w for the core sounds about right memory would be in the 100w range. Can you run something like Timespy just to verify your are getting ballpark performance for a 3090?


edit that was the 3090ti..... I guess it was 2x 8 to 12pin but still seems the issue is the memory. The 400w at idle also sounds like a bug your temps would be way higher lol.

It definitely has nothing to do with the wattage of your PSU My 850 prime TX in my secondary pc handles a 450w gpu no problem.

Guessing the card was used when purchased did you try contacting the original owner if that was the case?

I'm definitely not getting 3090 Performance as the GPU clock with not move from 210MHz under load.

The card was purchased used, Stock cooler was definitely never removed until I took it off, not that that means anything.

Would a bad waterblock install possibly cause excessive power draw at idle? I don't think it's misreporting, the card is throttling to 210Mhz because power draw is at MAX TDP+ at idle.

Only way I can get the GPU Core clock to go above 210MHz is by manually adding MHz in MSI AB app.

Edit: attached Time spy score and expanded GPU power readings from HWiNFO.

Looks like as if there is something wrong with the shunts for measuring current, if true, too soon after the card was new; definitely ask the original owner what was going on with the card.

Is there a a way I can visibly confirm this?

 

[oxrufiioxo]

Ouch.... Could just be a faulty card. You could try a bios flash but like with anything make sure you have a backup plan if that goes south. Anything can happen when modifying the cooling maybe the VRM isn't getting proper coverage but even then you would expect at least decent performance till the vrm thermal throttled. It still doesn't seem to be reporting properly in general though.

I don't typically modify cards but if I do I always make it a habit to thoroughly test the card in its out of box configuration for these kind of scenarios.

 

[pavle]

Visually the card's PCB would be different color somewhere close to the power connector (shunts look like/are larger resistors), but I would say first do anything to properly connect those two 8 pin connectors to your 12/16 pin adapter. All pins have to be connected; additional wattage comes from ground pins.

 

[HiResJunky]

Visually the card's PCB would be different color close to the power connector, but I would say first do anything to properly connect those two 8 pin connectors to your 12/16 pin. All pins have to be connected; additional wattage comes from ground pins.

So 8 pins on the GPU side needs 8 pins on the PSU side correct? I don't understand why the PCI-e cables supplied with my 850W power supply only have 6pins on the PSU side

 

[agent_x007]

Just FYI : You don't NEED 8-pin to PSU, if you have that 2-pin attachable part on 6-pin it will work just fine (connect those 2+6 parts together and put result of it into GPU's 8-pin).
But like was mentioned previously, one of those additional pins is designated as "sense" pin.
Which may take card in "fail safe" mode (locking max. turbo), if it isn't properly connected (or plug is damaged).
Make sure you connected ALL power to the card (ALL pins on power must be used), if it still throttles to safe mode regardless however - the only way to check shunts resistors for damage, is to unmount the cooler and check them manually/directly (it may not even be shunts, but some IC being damaged and reporting BS).

 

[Vya Domus]

Classic Nvidia power bug.

Not much you can do, it's unfixable with software, if possible I'd RMA, otherwise you need to look into doing a power shunt mod.

 

[HiResJunky]

Classic Nvidia power bug.

Not much you can do, it's unfixable with software, if possible I'd RMA, otherwise you need to look into doing a power shunt mod.

Don't think I can RMA it, I'm second owner. Guess they guy pulled a fast one on me. I will look into shunt mod.

 

[Vya Domus]

Don't think I can RMA it, I'm second owner. Guess they guy pulled a fast one on me. I will look into shunt mod.

By the way there is no guarantee that will fix it but it's worth trying. The reason this happens is because something in the power draw sensing circuitry is damaged, usually a resistor, it used to be with some older generation cards from Kepler era that you could just edit the bios around this power limit but with newer cards it can't be done as far as I know.

The shunt mod makes it so that it always reports a very low power usage figure hence fixing your problem with clocks.

 

[Psychoholic]

I assume changing clocks in afterburner does not help?
Have you tried setting max performance in the nvidia driver power options? Might allow 3d clock profile to run.

Off topic, but i got a geeky chuckle out of your typo (Intel 8086 oc'd to 5.2MHz)
Ironically the Original intel 8086 chip ran at 5Mhz

 

[skizzo]

So 8 pins on the GPU side needs 8 pins on the PSU side correct? I don't understand why the PCI-e cables supplied with my 850W power supply only have 6pins on the PSU side

the plugs on the PSU side do not need to match the plug shape/size/type on the other end. the way you're thinking is not how things work. For ex, take a moment to count how many pins are on the main motherboard power cable, there isn't 24 pins on the PSU side, yet there is 24 for your motherboard side....let alone, take a moment to realize that cable breaks out to two separate connectors on the PSU side yet it's a single connector on the motherboard side.

you DO need to pay attention to which end of the cable goes into the PSU though. normally the connectors are labeled very clearly. for ex should be something like "PSU" on the PSU end of the cable, and then "PCIE" on the end that would plug into the GPU. you wouldn't want to swap the ends around! though, I'm not sure if it's even possible, such as different pin outs and/or keyed shape of said pins

 

[HiResJunky]

the plugs on the PSU side do not need to match the plug shape/size/type on the other end. the way you're thinking is not how things work. For ex, take a moment to count how many pins are on the main motherboard power cable, there isn't 24 pins on the PSU side, yet there is 24 for your motherboard side....let alone, take a moment to realize that cable breaks out to two separate connectors on the PSU side yet it's a single connector on the motherboard side.

you DO need to pay attention to which end of the cable goes into the PSU though. normally the connectors are labeled very clearly. for ex should be something like "PSU" on the PSU end of the cable, and then "PCIE" on the end that would plug into the GPU. you wouldn't want to swap the ends around! though, I'm not sure if it's even possible, such as different pin outs and/or keyed shape of said pins

Got it, this helps alot thank you. I haven't kept up with the PSU and GPU power changes since 1080ti. Was grasping at straws to find a simple cause for this power issue.

By the way there is no guarantee that will fix it but it's worth trying. The reason this happens is because something in the power draw sensing circuitry is damaged, usually a resistor, it used to be with some older generation cards from Kepler era that you could just edit the bios around this power limit but with newer cards it can't be done as far as I know.

The shunt mod makes it so that it always reports a very low power usage figure hence fixing your problem with clocks.

I found a picture I took before installing the waterblock. This was as I cutting thermal pads. Anything noticable jumping out at you?

 

[pavle]

I marked what I believe are the shunts with yellow circle (you have to zoom in to see). Might be some other; try measuring resistance (will be very low, .something ohm, hard to measure). What a nasty situation...

 

[HiResJunky]

View attachment 291441
I marked what I believe are the shunts with yellow circle (you have to zoom in to see). Might be some other; try measuring resistance (will be very low, .something ohm, hard to measure). What a nasty situation...

Nasty indeed, this is the big downside to custom water-cooling, It's a huge deal to drain and take the card out. I'm trying to exhaust every other option. I'm definitely going to setup a test bench moving forward, can't rely on my good luck anymore.

Thanks for marking up the picture!

 

[Calenhad]

You did not test the GPU before stripping it? So you have no idea if you bought a DOA card or not?

 

[HiResJunky]

You did not test the GPU before stripping it? So you have no idea if you bought a DOA card or not?

Unfortunately this is correct, because I'm using a full water cooled setup I can't just plug in the card with air cooler to test it. Don't have a test bench setup

 

[80251]

Isn't the 8086 the OP is referring to a cherry picked 8700k?
On my Coolermaster v850-sfx-gold only 7 of the 8-pins from the PCIe 8+2 pin cables are connected at the PSU side, which is strange because the EPS 4+4 cable has all 8-pins connected on the PSU side.
So the person who sold the OP the card unsuccessfully shunt-modded it?

 

[HiResJunky]

Isn't the 8086 the OP is referring to a cherry picked 8700k?
On my Coolermaster v850-sfx-gold only 7 of the 8-pins from the PCIe 8+2 pin cables are connected at the PSU side, which is strange because the EPS 4+4 cable has all 8-pins connected on the PSU side.
So the person who sold the OP the card unsuccessfully shunt-modded it?

My CPU is a "special" edition 8700k. I will be removing the waterblock next week to examine the shunts but I don't think it was touched as the stock air-cooler had not been removed.

 

[Solaris17]

Looks like as if there is something wrong with the shunts for measuring current

This would be my first guess. When a power shunt is bad this is literally what happens. The card goes into protection.

 

[Vayra86]

Bottom end clocks with high reported power draw / perf cap Pwr on an Nvidia card tends to mean its bricked. The card defaults to a safe mode then where it locks to lowest power state at '100% tdp'

Exceptions do happen so if you can determine its just a faulty sensor might be lucky. I doubt it though.

 

[oxrufiioxo]

Bottom end clocks with high reported power draw / perf cap Pwr on an Nvidia card tends to mean its bricked. The card defaults to a safe mode then where it locks to lowest power state at '100% tdp'

Exceptions do happen so if you can determine its just a faulty sensor might be lucky. I doubt it though.

Even though it sucks for the OP to have to go through this the thread has been pretty informative to me. I've probably been hands on with 100-120 graphics cards over the last half decade and I've never actually seen this issue ever even with some very old GPUs.

At least 90% of the cards where Nvidia. Although out of all those cards I'm only aware of one dying to begin with or even having issues at all.

 

[Vayra86]

Classic Nvidia power bug.

Not much you can do, it's unfixable with software, if possible I'd RMA, otherwise you need to look into doing a power shunt mod.

Shunt mod can fix this? Nice

 

[80251]

So HiResJunky can't RMA? What if the seller gave him the original receipt? I've never seen an issue like this either. I once had a 1080ti fail on me, but it was because the VRM circuit wasn't designed to take the XOC VBIOS and I actually got it fixed.