GPU-Z reporting 7900XTX running at Gen 1.1

[Nostras]

Hi folks,

GPU-Z is reporting that my GPU is running at Gen 1.1 x16.

https://gpuz.techpowerup.com/23/03/22/tkz.png

That seems a bit unlikely, Adrenalin is stating that it's running fine at gen 4.

https://i.snipboard.io/yWTBxF.jpg

Is GPU-Z wrong or Adrenalin wrong?
Any way to confirm what speed the GPU is actually running at?

 

[GerKNG]

it's idling...

 

[Nostras]

it's idling...

https://i.snipboard.io/T16wZr.jpg

I tested that of course.

 

[pavle]

That render test is nothing for such a powerful card - please run something harder to display like a benchmark or demo (furmark is an extreme example) and report back.

 

[Nostras]

Ok, so I tried running superposition but it kept crashing on me with AMD screaming to me that it was running in safe mode (or something).
Perhaps that effectively means that the bandwidth is reduced to 1.1? Maybe it has been running that mode since boot.
After a reboot it now properly says 4.0 again.
If I catch it again I'll try to test it, but I think it's related to that.

 

[floofyhunters]

It switches to 1.1 to save on power

 

[Naki]

* Please fill in your PC specs in your forum profile.

* What motherboard & CPU do you use?

I think there is way to enable Safe mode, if holding a certain keyboard key during Windows OS boot-up.
So, maybe inadvertently that is what happened here.
OR, if your keyboard is old, maybe that same key got "stuck"?

 

[Nostras]

* Please fill in your PC specs in your forum profile.

* What motherboard & CPU do you use?

I think there is way to enable Safe mode, if holding a certain keyboard key during Windows OS boot-up.
So, maybe inadvertently that is what happened here.
OR, if your keyboard is old, maybe that same key got "stuck"?

7950X3D + X670E-I.
And no, not Windows running in safe mode. I got a popup from Adrenalin telling me an error was detected and that the driver was running in safe mode.
Not windows safe mode, whatever AMD says their safe mode is.
I would've most definitely noticed it if Windows was running in safe mode lol.

Edit: I didn't really think it was necessary to screenshot it as I thought it would've been easy to find, but surprisingly it is not.
It's this one.

https://preview.redd.it/amd-driver-crashes-again-what-should-i-do-now-p-s-it-v0-7wbhwsl33ek81.png?width=640&crop=smart&auto=webp&s=0812d7cb9f2af367ccec95e26054cd8be82b9031

 

[Calenhad]

Have you tried re-seating the GPU?
Are you using a riser cable, or is it plugged directly into the motherboard?

 

[Nostras]

So it happened again. No funny business with safe mode or anything.

https://i.snipboard.io/M963qn.jpg

I'm not really looking for a solution, if the GPU is truly running at 1.1 mode I either need to check my setup or scream at ASUS.
I'm looking for confirmation that the card is truly running at 1.1 mode, so any suggestions to confirm this are welcome.

And yes, I'm using a Meshlicious with a gen 4 riser.

 

[Naki]

Obvious suggestion would be - try without any riser.
If not possible at your place, maybe using a friend's PC for the test?

And please do fill in your PC specs.

 

[Nostras]

Obvious suggestion would be - try without any riser.
If not possible at your place, maybe using a friend's PC for the test?

And please do fill in your PC specs.

My first things to try would be to force gen 4 in the bios instead of gen 1 to check if auto is misbehaving, but if that doesn't work I'll have to clean the riser and refit it.
Not using the riser is not really possible as the PSU blocks the GPU in an ITX Meshlicious build.

7950X3D
ASUS X670E-I
KF556C40BBK2-64
Hellhound 7900XTX
SF750
Meshlicious Gen 4 edition
EKWB EK-AIO 280 D-RGB
SN850X
SX8200 Pro
660p

 

[Tropick]

And you're 100% certain you have the PCIe 4.0 version of this riser? It looks like there's both a 3.0 and 4.0 version listed together on the site and even the 4.0 version has only been validated with RTX 3000 and RX 6000 GPUs. This sounds like a PCIe link negotiation issue.

 

[Nostras]

And you're 100% certain you have the PCIe 4.0 version of this riser? It looks like there's both a 3.0 and 4.0 version listed together on the site and even the 4.0 version has only been validated with RTX 3000 and RX 6000 GPUs. This sounds like a PCIe link negotiation issue.

Yes, I bought the gen 4 edition.

https://www.caseking.de/ssupd-meshlicious-full-mesh-pcie-4.0-edition-mini-itx-gehaeuse-weiss-gesp-012.html?tplview=desktop

But look at this picture after I forced gen 4 in the bios.

https://i.snipboard.io/YUfwmA.jpg

Can you guys spot the performance difference?
I'm just not sure if the reboot did the trick or forcing gen 4 did it.

I guess that's one way to test it. Seems like GPU-Z was right.

 

[Nostras]

Well folks, it's been a couple days and this issue still remains.
I did do some overclocking where I slightly bumped the BCLK to 101.5MHz for a small overclock.
I thought that could perhaps have been the culprit, but even after resetting that to 100MHz this problem still occurs.

Notably the device seems to always boots with the GPU in Gen 4 mode but regresses to Gen 1 after an unknown amount of time.
I've yet to see a pattern here, but it does not seem to be idle related as I saw it happening while I had 2 android emulators and a twitch stream open.
I did try removing power saving in the windows energy plan but that did nothing sadly.

I've also removed the GPU and riser and tried cleaning + refitting but that seemed to have had no effect (so far).

The next step is to just remove the motherboard from the case and use the PC testbench style.
I really dislike doing this as my cat may try something funny, but I'm frankly out of ideas.

If the problem resolves I'm contacting SSUPD for a replacement riser.
If the problem persists I'm contacting ASUS for support.

If anyone has some interesting thoughts on this I'm all ears.
A confirmation that a bad riser could cause this would be nice for example.
I just figured a bad riser would just cause stability issues or a regression to Gen 3, not Gen 1.
Gen 3 would've been poor but nowhere near as notable if notable at all.

I'm just jotting down my thoughts for anyone else that finds this topic and can see similar problems (and hopefully at the end see a solution).

 

[Calenhad]

PCIe communication will regress as far as it needs to if you have communication errors. A bad riser will cause the bus to go gen 4 -> 3 -> 2 -> 1.1 in search of stability. Worst case it will crash if 1.1 speed is still unstable.

Have you tried without the riser at all?

 

[Nostras]

PCIe communication will regress as far as it needs to if you have communication errors. A bad riser will cause the bus to go gen 4 -> 3 -> 2 -> 1.1 in search of stability. Worst case it will crash if 1.1 speed is still unstable.

Have you tried without the riser at all?

Haven't seen it running at 3 or 2.
And no, I've yet to do that as it's a pain. I'll probably do it some time tomorrow.

 

[Calenhad]

Haven't seen it running at 3 or 2.

Yeah you won't either. My guesstimate is that this correction happen "instantly" when the pile of errors cause the drop to 1.1.

 

[Nostras]

Yeah you won't either. My guesstimate is that this correction happen "instantly" when the pile of errors cause the drop to 1.1.

Can you like see this somewhere. I tried event viewer but after a cursory glance I gave up on that.

 

[Calenhad]

Can you like see this somewhere. I tried event viewer but after a cursory glance I gave up on that.

I am not aware of any easy way to log this. I assume it is possible to get the driver into debug mode to do it. But I have never looked further than understanding bus communication/error checking. Probably 100x easier to do this kind of logging in Linux as well.

Either way that riser is the prime suspect. You really need to either test without it or test another one.

 

[Mr.Scott]

Change windows power plan to high performance.

 

[Nostras]

Change windows power plan to high performance.

Doesn't this bork AMD's sauce with the 7950X3D?

 

[Calenhad]

Doesn't this bork AMD's sauce with the 7950X3D?

Would not be a permanent fix. You can do it for testing purposes.

But quite honestly I would say get that riser checked out first thing. Stop looking for workarounds.

 

[Nostras]

Well this was a chore.

Now let's see if it happens again.

 

[Tropick]

Well this was a chore.

View attachment 290247
Now let's see if it happens again.

Hey a solid diagnosis is worth a few hours of disassembly in my book, especially when the patient is a 7900XTX. Keep us updated