RX 9070 XT freezing/locking up only on desktop, anyone else?

[RejZoR]

I have this weird issue with it that I just can't seem to figure out and it's driving me crazy. I can game for hours from mild load of Overwatch 2 to heavy load of Oblivion Remastered or Doom The Dark Ages and it will never have any problems. But on desktop, what I usually do is have Youtube or other streaming video in PiP (Picture in Picture) and do other things like edit photos or browse web and suddenly, system would just lockup/freeze, sound would turn into static noise loop and this would go on for up to 30 seconds after which system unfreezes and everything is back in order like nothing happened. GPU crash detection is not triggered, there is nothing in Windows Event logs other than network reinitializing yellow exclamation trinagles after the lockup/freeze. I thought Firefox was causing this, but after switching to Vivaldi for several days and it just happened again. I just don't get it.

I've tested with everything stock, CPU, RAM, graphic card, still happening. Done memtest on RAM, all fine. I've removed my AE-9 dedicated soundcard and ran onboard audio for days and it still happened. I can't think of anything else being the issue other than graphic card which is the most "recent" addition and I've never had such issue with RTX 3080 prior (I've done clean system reinstall for Radeon). And the weird thing is, heavy load in games is not an issue where I could potentially blame PSU for it. But on desktop with minimal load. How does that make any sense?! I've also gone through several drivers, but I really can't go back to release drivers where I don't remember the issues, but they were bad for everything else, especially Radeon Chill and AFMF.

Any other idea because this is really stupid and it's annoying me so much.

 

[Kwadratowicz]

Just guess, but if You don't have set power setting for maximum performance try that.
And Do You have monitor setup with hdmi or dp?, Cause I've had problems with sound while using hdmi. Now on DP there is no problems with sound.

 

[Event Horizon]

I can only offer random ideas as I don't have this issue on my 7800X3D + 9070. Switch PCIe speed in BIOS from auto to manual. Disable hardware acceleration in the browser. If the issue is fast/easy to reproduce, monitor or log the GPU clocks at the moment it freezes. The network reinitialise part seems very strange, maybe dig deeper into that if possible. If all else fails and it's quick and easy to reproduce then filing a bug report might be your best bet. There aren't that many Radeon users out there, so AMD could be starved for quality bug reports relative to Nvidia.

 

[Fishymachine]

I'd try checking with a new PSU (900 or 1000W) to see if it's some transient spike

 

[Peganeo]

I have same issue and I have an fix for that. It happening only hardware-accelerated apps. Download mpogpufix and do it my settings, then reboot the computer and you will see it's fixed.

Releases · RedDot-3ND7355/MPO-GPU-FIX

MPO Fix for 5000/6000/7000 AMD GPUs & RTX 2000/3000/4000/GTX 1600 - RedDot-3ND7355/MPO-GPU-FIX

github.com

 

[Kwadratowicz]

I have same issue and I have an fix for that. It happening only hardware-accelerated apps. Download mpogpufix and do it my settings, then reboot the computer and you will see it's fixed.
View attachment 406520

Releases · RedDot-3ND7355/MPO-GPU-FIX

MPO Fix for 5000/6000/7000 AMD GPUs & RTX 2000/3000/4000/GTX 1600 - RedDot-3ND7355/MPO-GPU-FIX

github.com

I must try it Myself, cause I've got problems with alt tab from games.

 

[Bomby569]

I'd try checking with a new PSU (900 or 1000W) to see if it's some transient spike

transient spikes on desktop?

boy did i ever saw this movie with my 5700xt. Try reverting to older drivers and good luck with that

 

[RejZoR]

I'd try checking with a new PSU (900 or 1000W) to see if it's some transient spike

Transient spike? Why would there be one on desktop doing nothing special and not have the issue with it when I'm doing image upscaling that constantly alternates with heavy GPU load and heavy CPU load, first GPU to upscale image and then CPU when it's encoding it to PNG? I can process 50 images and no issues. But then it just randomly locks up during Youtube in PiP on desktop... Also, if that was the problem, I'd expect one in games because I'm using Radeon Chill which causes GPU clock to go up and down dramatically based on how Chill is limiting framerate and I've not had issues in any game. Also I'm using PSU that I'd consider high quality enough to not be affected by that. My PSU also has a switch between multirail and single rail power delivery and it didn't change the situation.

I've actually had system lock up issues years ago because of faulty DisplayPort cable. But I've replaced it and the issue is still there so it's not DP issue this time.

I've done a total clean Windows reinstall with minimal tweaking to see if I had some stupid software conflict somehow though I doubt it.

I remember having issues with ULMP literal decades ago on some Radeon card, I'll try the MPO GPU Fix from above if I still get the issue.

 

[Bomby569]

Transient spike? Why would there be one on desktop doing nothing special and not have the issue with it when I'm doing image upscaling that constantly alternates with heavy GPU load and heavy CPU load, first GPU to upscale image and then CPU when it's encoding it to PNG? I can process 50 images and no issues. But then it just randomly locks up during Youtube in PiP on desktop... Also, if that was the problem, I'd expect one in games because I'm using Radeon Chill which causes GPU clock to go up and down dramatically based on how Chill is limiting framerate and I've not had issues in any game. Also I'm using PSU that I'd consider high quality enough to not be affected by that. My PSU also has a switch between multirail and single rail power delivery and it didn't change the situation.

I've actually had system lock up issues years ago because of faulty DisplayPort cable. But I've replaced it and the issue is still there so it's not DP issue this time.

I've done a total clean Windows reinstall with minimal tweaking to see if I had some stupid software conflict somehow though I doubt it.

I remember having issues with ULMP literal decades ago on some Radeon card, I'll try the MPO GPU Fix from above if I still get the issue.

the exact same thing happened with so many of us with the 5700xt, and the answers were the same. You got to tune out the most absurd, because they will blame everything, even when it makes no sense at all.

try old drivers, i bet they will fix it. At one point we were using 2 year old drivers

 

[RejZoR]

the exact same thing happened with so many of us with the 5700xt, and the answers were the same. You got to tune out the most absurd, because they will blame everything, even when it makes no sense at all.

try old drivers, i bet they will fix it. At one point we were using 2 year old drivers

I mean, RX 9070 XT is not even 1 year old and the release drivers were rough in regards to Radeon Chill. I can't go "back" with drivers.

 

[Dr. Dro]

I have same issue and I have an fix for that. It happening only hardware-accelerated apps. Download mpogpufix and do it my settings, then reboot the computer and you will see it's fixed.
View attachment 406520

Releases · RedDot-3ND7355/MPO-GPU-FIX

MPO Fix for 5000/6000/7000 AMD GPUs & RTX 2000/3000/4000/GTX 1600 - RedDot-3ND7355/MPO-GPU-FIX

github.com

I must add that disabling MPOs is not a one-stop-shop fix without consequences. Even as a last resort, it's something you should avoid if it can be helped. To simplify, doing so will cause your game's swapchain and any other windows, as well as the desktop to run through the DWM compositor, that will manage a single scanout instance. This will incur heavy penalties on render latency, frame timing and frame rates overall. Ideally, you want MPOs enabled, and for your games to be running with independent flip, something that is not possible with MPOs off. Please refer to:

Presentation Model (D3D11-12)

For when you wonder why things are the way they are...

wiki.special-k.info

It should never be disabled in a modern computer, and in fact, Windows 11 24H2 ignores the registry key setting to disable it. Demand driver bug fixes from your graphics vendor instead. I'm not saying blame AMD exclusively because MPO support is spotty on Nvidia too, and there's been a hard push for this (and for good reason, it's what enables smooth gaming). If you run into issues that demand MPOs to be disabled, make sure to report the bug through the appropriate channels.

I mean, RX 9070 XT is not even 1 year old and the release drivers were rough in regards to Radeon Chill. I can't go "back" with drivers.

Rolling back drivers will likely not help your case. What it sounds like to me is that you're probably running into that classic bug with the AMD drivers that cause them to TDR/freeze whenever a new media decoding context comes up. They just can't seem to get rid of it. This may sound very silly, but try disabling hardware acceleration on your web browser for the next few days. If it's resolved, then you know what you're running into.

 

[Peganeo]

I must add that disabling MPOs is not a one-stop-shop fix without consequences. Even as a last resort, it's something you should avoid if it can be helped. To simplify, doing so will cause your game's swapchain and any other windows, as well as the desktop to run through the DWM compositor, that will manage a single scanout instance. This will incur heavy penalties on render latency, frame timing and frame rates overall. Ideally, you want MPOs enabled, and for your games to be running with independent flip, something that is not possible with MPOs off. Please refer to:

Presentation Model (D3D11-12)

For when you wonder why things are the way they are...

wiki.special-k.info

It should never be disabled in a modern computer, and in fact, Windows 11 24H2 ignores the registry key setting to disable it. Demand driver bug fixes from your graphics vendor instead. I'm not saying blame AMD exclusively because MPO support is spotty on Nvidia too, and there's been a hard push for this (and for good reason, it's what enables smooth gaming). If you run into issues that demand MPOs to be disabled, make sure to report the bug through the appropriate channels.

I didn't know that MPO was so important, thank you for explaining. However, this is the only way I can solve this problem on my computer. If there is a way to solve it without turning off MPO, please let me know. Also I have freezing issues when playing games. I tried many things and still doing that.

 

[texh]

As to replicate the problem, use the old gpu, if no crash then def the new gpu. RMA time....

 

[Bomby569]

I mean, RX 9070 XT is not even 1 year old and the release drivers were rough in regards to Radeon Chill. I can't go "back" with drivers.

I guess it's even worst this time :/

 

[io-waiter]

Tbh it sounds like the GPU struggles at low voltage and light load scenario. Since there is no OC, bad GPU?
One way to troubleshoot would be to not let the card drop down to to low voltage, iirc igorslab has some tool for this, if the issue is gone then the GPU is likely the culprit, otherwise it would be a driver/sw at fault..

 

[Bomby569]

Tbh it sounds like the GPU struggles at low voltage and light load scenario. Since there is no OC, bad GPU?
One way to troubleshoot would be to not let the card drop down to to low voltage, iirc igorslab has some tool for this, if the issue is gone then the GPU is likely the culprit, otherwise it would be a driver/sw at fault..

if that was the real issue it would happen all the time when the load is low.

 

[RejZoR]

Disabled HW acceleration in Firefox and this time it seemed like system wanted to freeze again, but it just happened for a fraction of a second and returned back to normal in same fraction of a second. What's odd is that AMD software detected GPU crash, but didn't even reset my undervolt which is what it usually does when crash is detected (also yes, I've kept GPU default and it did anyway so undervolt isn't an issue).

It was again the Paint.NET and Firefox with video in PiP combo. I'll now disable Paint.NET HW acceleration of interface as I can't do that for effects as they are so slow it's unusable. So god damn annoying all this...

 

[Frick]

Disabled HW acceleration in Firefox and this time it seemed like system wanted to freeze again, but it just happened for a fraction of a second and returned back to normal in same fraction of a second. What's odd is that AMD software detected GPU crash, but didn't even reset my undervolt which is what it usually does when crash is detected (also yes, I've kept GPU default and it did anyway so undervolt isn't an issue).

It was again the Paint.NET and Firefox with video in PiP combo. I'll now disable Paint.NET HW acceleration of interface as I can't do that for effects as they are so slow it's unusable. So god damn annoying all this...

My undervolted 6950xt was game/stress test stable but not HW accelerated Paint.net stable, and that was when I stopped messing around with undervolting.

 

[io-waiter]

if that was the real issue it would happen all the time when the load is low.

Not neccesary, it just takes a single "point" on the voltage curve to be overly optimistic before things start go wrong.
It can happen when you undervolt (as @Frick mentioned above), but if the GPU had bad QC it is possible it has a bad spot on its "curve" for some reason.
It would be pretty easy to rule out by locking the GPU to a known good section of the voltage curve.

EDIT: Clarification, it's not just about voltage and frequency, it's also about what load, or task that the GPU is subjected to under those circumstances. With this said I'm not arguing to rule out the possibility of a SW error, just that rulig out a HW error would minimize the scope of the investigation for the root cause.

 

[Chrispy_]

My undervolted 6950xt was game/stress test stable but not HW accelerated Paint.net stable, and that was when I stopped messing around with undervolting.

That's crazy, but not actually unbelievable. Some of the smallest, simplest things can cause the worst problems!

I remember Amazon's "new world" game bricked a bunch of RTX 3080 and 3090 graphics cards because there was some code that nobody had thought to sanity-check and it essentially ran the GPU at tens of thousands of FPS in a power-virus way that circumvented all of Nvidia's driver checks for power virus and eluded the throttling and power draw protections in the driver.

I wouldn't be surprised if something simple and lightweight caused havoc by simply asking for the impossible, because nobody has actually debugged your exact hardware/software scenario.

 

[c2DDragon]

The only issue I had was the GPU wasn't fully used in like 50% of the heavy games I play.
I thought it was due to the drivers, I guess it is but I disabled HAGS in Windows settings and the problem is now solved.
Did you try to disable HAGS and Ulps ? (Ulps : https://www.sevenforums.com/tutorials/316913-ulps-ultra-low-power-state-disable-amd-crossfirex.html)

Also, did you disable dynamic refresh rate in Windows settings ?


Finally, does GPU-Z reports that your graphic card is using the PCIe port at x16 5.0 (PCIe gen 5) ?

 

[Luminescent]

Check windows event viewer and see what triggered the crash

 

[texh]

OP you can use diff pcie port, some cases its RAM related too.

 

[Bomby569]

Not neccesary, it just takes a single "point" on the voltage curve to be overly optimistic before things start go wrong.
It can happen when you undervolt (as @Frick mentioned above), but if the GPU had bad QC it is possible it has a bad spot on its "curve" for some reason.
It would be pretty easy to rule out by locking the GPU to a known good section of the voltage curve.

EDIT: Clarification, it's not just about voltage and frequency, it's also about what load, or task that the GPU is subjected to under those circumstances. With this said I'm not arguing to rule out the possibility of a SW error, just that rulig out a HW error would minimize the scope of the investigation for the root cause.

what voltage curve, when you are just basically idle? that's the flat part of the curve no, or better no curve at all.

 

[io-waiter]

what voltage curve, when you are just basically idle? that's the flat part of the curve no, or better no curve at all.

Well yes, depending on how much load that light load is, and how the flat part handles it? Or am I missing here?