Question Strange issue with GPU ?

[Chasenuva1]

I had an old system that i upgraded recently.
old system worked great and had these specs:
i9-9900kf, 16gb ddr4, ROG 4090, Taichi Z390 mobo, 2x nvme drives

New system specs:

MBD: ROG Z790
CPU: i9-14900kf,
GPU: ROG 4090
SSD: 2 x NVMe
RAM: 64GB DDR5

I swapped out all the components from the old system to the new system as shown ^
Immediately it wouldn't boot and after a significant search online the solution was to change some settings in bios related to the nvme drives. That's not the current problem, just want to include the whole story. Maybe it'll be of significance.

Once i got everything to boot, i haven't been able to keep the system running for more than about 10 minutes due to system stability issues.
It is important to note that these issues persist to a new OS install on a third nvme drive, even if i remove the other two nvme drives.
With nothing running except the Desktop, moving the mouse around shows the frame rate is about 5fps.
Task Manager shows 100% GPU usage at all times. GPU is used mostly (65-98%) by a generic windows process related to graphics: Client Server Runtime Process
BSODs with many and varying stop codes, mostly related to graphics drivers but not all of them are.
About 25% of the time is a boot-to-black-screen where the system doesn't even finish posting.

Drivers/Firmware has all been updated (gpu, mobo, cpu, etc), no change.
OS has been reinstalled, no change.
Interestingly, when the graphics drivers are completely uninstalled and before Windows Update installs it's default drivers, I see no issues.
Replacing the 4090 with a working 3090 i had in my closet fixes the problem immediately. Problem returns when the 4090 is reinstalled.
Setting up the old system and putting the 4090 in it now doesn't work, issues are the same.

So i concluded that it was a gpu issue. Put in an RMA to ASUS, and they replaced it with a new replacement 4090.
Strangely, problem persists. Which points me to it not being a gpu issue.
However, while waiting for the 4090 RMA to process, the 3090 has been running the system with zero problems.

I'm out of ideas on how to proceed. Everything points to the 4090 being the problem, but two different 4090s have the same issue.
Any suggestions?

 

[Hotrod2go]

Are you still using the same PSU from the old system?

 

[thestryker]

What is the exact power supply?

Is the video card directly connected to PCIe socket or is there a riser cable?

Are you using an adapter to connect the 4090 to PCIe power cables?

Have you tried booting into safe mode to see if the system happens to be stable there?
(this would point towards some sort of software issue)

 

[rene641]

Do you have latest BIOS? XMP mode on? Resizable BAR enabled?

 

[Eximo]

64GB of DDR5 at what speed, what configuration?

 

[Chasenuva1]

Are you still using the same PSU from the old system?

What is the exact power supply?

Is the video card directly connected to PCIe socket or is there a riser cable?

Are you using an adapter to connect the 4090 to PCIe power cables?

Have you tried booting into safe mode to see if the system happens to be stable there?
(this would point towards some sort of software issue)

no, it's a new psu. however, issue persists if i swap back to the old psu. (old one worked great with the 4090 before i did the upgrade)
new: MSI MEG Ai1000P
old: EVGA SuperNOVA 850GT

i have a riser cable. I am unable to test without the riser cable due to case design and 4090 size. (no, i don't want to do a caseless thing either cuz i don't want to break the 4090 putting all that weight/torque only on the connector)

the old system used the included 4x 6+2 PCIe to 16 pin 4090 plug with only 3x of the 6+2 PCIe power cables connected (psu didn't have a 4th slot and i didn't want to overload one of the rails)
the new system has a 16 pin 4090 style plug, which i tried first but when that didn't work i swapped back to the 3x 6+2 PCIes and when that also didn't work i tried 4x 6+2 PCIes because the new psu had a 4th rail for it.

safe mode has the same issue. (well, this was with the card before i sent it in for RMA. i'll try again with the replacement tonight)

Do you have latest BIOS? XMP mode on? Resizable BAR enabled?

yes, yes and no (tried both), and what's a ReBAR? (lol that acronym)

64GB of DDR5 at what speed, what configuration?

DDR5 6400
not sure what you mean by configuration, but the timings are 32,39,39,102.
unless you mean what slots in the mobo... then it'll be the A1 and B1 slots.

 

[thestryker]

i have a riser cable. I am unable to test without the riser cable due to case design and 4090 size. (no, i don't want to do a caseless thing either cuz i don't want to break the 4090 putting all that weight/torque only on the connector)

Try setting the PCIe slot to 3.0 and see what happens.

 

[rene641]

Was the rtx 3090 connected via a riser as well? I have no experience with the riser, but ASUS writes https://www.asus.com/support/faq/1041931/

Resize BAR
https://www.asus.com/global/support/faq/1046107/

 

[Chasenuva1]

Was the rtx 3090 connected via a riser as well? I have no experience with the riser, but ASUS writes https://www.asus.com/support/faq/1041931/

Resize BAR
https://www.asus.com/global/support/faq/1046107/

why do the links you provided go to a different site than the one the text shows? -__-
That's fishy as hell, but I should test it before assuming you did it on purpose.
https://www.asus.com/global/support/faq/1046107/
Yea, mine didn't get changed. You're fishy. I don't trust you anymore.

Try setting the PCIe slot to 3.0 and see what happens.

will try and will edit this post in a few minutes.

Edit: Yup that did it! Amazing! Thank you!
So does this mean my riser is too old or something? will a new one allow me to run gen5? Ugh, that's so expensive for a short cable......

 

[thestryker]

will try and will edit this post in a few minutes.

Edit: Yup that did it! Amazing! Thank you!
So does this mean my riser is too old or something? will a new one allow me to run gen5? Ugh, that's so expensive for a short cable......

My assumption is that your cable didn't have enough signal integrity for PCIe 4.0, but the other cards you were using didn't utilize enough bandwidth to expose it. The 4090 is the only card I'm aware of with a noticeable, though still low, performance loss at PCIe 3.0 x16 bandwidth. You'd definitely need a new riser cable to run at 4.0 since this resolved it. I'd suggest searching for what specific models people who have 4090s have used successfully as it might not take a super expensive one.

I wouldn't bother trying to get a PCIe 5.0 riser as any likely worth buying would cost way more than is reasonable.

 

[rene641]

why do the links you provided go to a different site than the one the text shows? -__-
That's fishy as hell, but I should test it before assuming you did it on purpose.
https://www.asus.com/global/support/faq/1046107/
Yea, mine didn't get changed. You're fishy. I don't trust you anymore.

Do not swear at anyone and choose your words carefully. I put the links normally on the Tomshardware website and I am not responsible for what happened to them. This should be detected by the Tomshardware administrator. I expect an apology from you...