PC Drivers & Software

[radu1006]

For some reason my 7900xt since the 24.x.x drivers were introduced, has a very low idle voltage. Also at the same time with these drivers , I started having Crashes and black screens in idle.

Can this be related? It Sometimes drops to 59mV. I suspect that it my drop even lower and when there is a spike for a task, the driver will crash.

Does anyone know if this voltage is normal? Also even when setting a minimum voltage for the GPU clock does not seem to affect the clock in idle.

I want to mention that I have disabled ULPS in the registry, and it seemed to be stable for about a week, but then I started having these instability issues again.

This voltage drop has to be driver related, since if I revert to 23.12.1 drivers voltage never goes that low. It always stays above 100mV.

[FunkZ]

When the 7900XT/XTX first launched users complained about high idle voltage and board power. AMD made adjustments to the drivers and idle power usage improved. (decreased)

Your 7900XT looks normal to me. I'm running 24.8.1

[radu1006]

Do you also use HDR? Because HDR uses more power usually. Nvidia GPUs in idle use about 40-50W with HDR enabled.

Also had 4 crashes with driver timeout in the last 2 days. If I revert to 23.12.1 the driver that has a higher idle power usage, I don't have these issues.

[radu1006]

My theory so far is that because of the low voltage in idle (aggressive power saving), when a more intensive background task like an update starts, the display driver crashes.

Right now I am trying to force a bit of GPU usage with Wallpapre Engine to see if it makes a difference. Unfortunately the min clock frequency setting from Adrenalin does not do anything on desktop/browser. Or maybe it does not work at all.

[FunkZ]

The screenshot I posted is with no other applications open. As soon as I open something else, even a browser window, voltage and power usage go up. I do not get crashes or black screens.

[radu1006]

Indeed there is not much load, but I have the browser open and 8 apps open in system tray as well.

The problems that I get are if I leave the PC in idle for a while. When in idle, some apps: like browsers, antivirus, even windows will start doing updates in the background.

If the voltage is very low at that time, the update task will do a spike and there is a chance that the GPU will not respond fast enough and high power is requested a low voltage and the driver crashes. It is similar behavior, as for undervolting.

AMD seems to have the voltage set borderline low.

You may not experience these crashes, as it also depends on the silicon lottery and maybe your GPU is a bit more stable at these low voltages, but a lot of users complained about these driver timeouts.

AMD should up the voltage to 100-200mV to improve stability. This will only add 1-2W of extra power in idle. Under load it would not matter.

[radu1006 · Accepted solution]

For the last few days I have been testing with the Metrics Overlay enabled on desktop. This seems to increase the voltage in idle at around 600mV.

It was quite stable, but I will give it a few more days. This increases the power usage in idle to around 20-30W. For me this is not a problem, but it is probable that it may be stable even with lower voltage, maybe around 200-300mV.

Anyway, this is an ugly workaround to have the overlay displayed all the time.

AMD, please check this problem and increase the idle voltage if needed to solve this issue!

I have already opened a bug about this a while ago.

[radu1006]

Tested with 24.10.1 and it has the same problem with low voltage. The Overlay workaround still works fine.

Adding this reply, so that the people will still see the post and workaround.

[GreenCobra]

I have a similar problem with the 24.10.1 drivers. Every time at idle the drivers crash and in task manager i don't see the 7900xt anymore, only the integrated graphics. To get it to work i have to restart the PC.

The voltage doesn't seem to drop from 680 at idle however your solution with the metrics overlay works and the crashes stop.

The crashes also stop if i have the adrenaline software open.

Edit: Sorry the voltage does drop to 55. I used adrenalin to see the voltage but it maintains at 680 thats why no crash with it open as i said above. I now used GPU Z and i see it drop to 55.

[radu1006]

Can we have this post pinned? So that the solution is more visible?

Thank you

[PlumBoniT0]

 @radu1006  OP, Can you try something out? If possible of course. Install 23.1.1 as driver only. no Adrenalin.

I have a wild theory here, since 7000 series had high idle power consumption, and lots of reviewers mentioned it when it released, AMD had to mess with the power states of these cards and optimize it's idle power consumption further.

If you check the driver notes, you'll see "high idle power consumption" as a known issue for multiple driver releases. Until 23.7.1 where they marked high idle issues as fixed for single monitor, and on 23.8.1 they noted down multi-monitor issues as finally fixed. that's eight months of updates.

By messing with the power settings to solve the "high idle power consumption", they rendered some of these cards unstable at idle, since obviously they couldn't test them all. Why some cards fail and some don't? I have no idea. I'm assuming a bit lesser silicon, or maybe related to the OC card models. Unsure at this point. 

If you test this and it works without having to use the overlay to increase idle power consumption, we could probably set up a nice bug report for AMD to look into and maybe fix this once and for all.

[radu1006]

It is not a theory it is actually what happened. But AMD messed with the power settings starting with 24.1.1. This was the first driver to cause instability in idle. 23.12.1 was working fine in this situations and one solution is o use this driver, but you will be missing the support for newer games.

Note: this only applies to these idle crashes, not crashes while playing games.

I have opened bug reports with almost every new driver version, but AMD is ignoring them. Probably if more people would be doing this they will notice.

The problem is that most people open bug reports that do not have much details and AMD probably cannot reproduce the issue. Also other people are happy that there is a workaround and do not open bug reports about this.

[PlumBoniT0]

Another observation that I noticed, with my XTX and monitor setup (Samsung G70A 4K 144hz), minimum board power consumption on GPUz when using DisplayPort is 12W (after a day of use).

When using an HDMI cable, the power consumption rises to 20W minimum, and throughout all these months, I've found the GPU is absolutely more stable when I'm using HDMI.

And yes, just like you mentioned, this is separate from the (lots of) users that are getting timeouts and such while gaming, this is really an idle state crash because seemingly AMD is pushing the power savings a bit too much which renders some cards unstable.

This is confirmed since none of our GPUs crash under load, hell I haven't gotten a single timeout or artifact while gaming. Nothing at all, absolutely zero issues when the GPU is under load.

I'll try to gather some more info and maybe try to ask more users that have these symptoms to submit more bug reports. It is absolutely the idle power states being too low what's causing all of this. 

[PlumBoniT0]

I have a question @radu1006 , which PSU are you currently using with the system? 

Do you by any chance have a PSU manufactured before the card was made? or just a model that was released previously. 

I am sure this is power related at this point, but I've also never considered it being my PSU. and PSUs have changed a lot in the last 2 years or so.

Just curious if you're running an ATX 3.0 / 3.1 PSU already, or (much like most other users having this issue), an older but quality unit. 

[radu1006]

I am using a Seasonic modular GM12 850W PSU. It is a pretty new model. Before that, I used a B12 650W and had the same issues.

Since older drivers did not have this issue, I would dismiss the PSU as a cause.

Most probable cause is the low voltage solution in order to lower the power consumption. I would expect some GPUs to behave a bit better, since the silicon lottery can make some chips a bit higher quality.

But I am pretty sure that the main cause is the low voltage. Also remember that nvidia found another way to lower the consumption without lowering the voltage.

[PlumBoniT0]

Gotcha, it's so odd that this is the case. I had that theory when I posted that message but I've now changed my PSU to a modern unit and my GPU is still acting up at idle....I've changed most of my system now, the only thing left to swap is the monitor (I don't have another high refresh rate monitor around)

It's odd because it's happening to a couple dozen users online at least, and if it really were a bad component that's unstable at low voltages, it's something that's happening to a lot of users

I also managed to improve stability at idle by raising power consumption, however my GPU still crashes on startups sometimes.

I still think this could be somewhat monitor related, but until I can get myself a new monitor I can't confirm that.

[PlumBoniT0]

I have some news regarding the instability at idle. Can you try running the latest AMD driver 25.3.1, without the overlay? If you still have the card. They have changed stuff regarding how the cards idle. It seems to have fixed mine, or at least made it less prone to crash. can you try yours? 

[radu1006]

Well, it does not seem to have changed much with 25.3.1. It took 22 mins after I disabled the overlay to get a freeze, then black screen after a few seconds and I had to hard reset the pc.

Had a driver timeout and recover in the event viewer afterwards.,

[PlumBoniT0]

Ok darn, was worth a shot. It seems like your card might be even more unstable than mine is in terms of voltage floor. 

On my card at least, I've discovered this is 100% related to memory voltage, or memory related to some extent. I'd be curious to know if yours is. I have also filed a report for my card's board partner and they seem to be looking into this more seriously. If AMD is not going to listen to users, I hope they at least listen to board partners, or that the board partner can release a bios update for the card with higher voltages or no power-saving states.

Since the overlay is not a solution for me as I  don't use Adrenalin, I have built a bunch of scripts that load up the core and prevent it from going into zero-rpm mode, and one that keeps the memory from idling, but allows the core to idle. That last one keeps the system completely stable.

With this latest bios (25.3.1) they changed some stuff that improved stability on my card, it doesn't seem to crash in regular use anymore, only when heavily idling, but it will still crash. Just not as often as before. Nothing else has changed.
 
I'll keep you posted if anything comes up from this. I'll keep pushing because it's such a simple problem that has a super simple solution and it still hasn't been acknowledged. It's asinine to me how the card will game just fine but crash at idle, and all it needs is a toggle on Adrenalin or a bios update that doesn't let the card idle as hard as it does.

[radu1006]

They had to mess around with the voltage in the drivers and make them unstable.

I am pretty sure that this problem will also resurface on the 9070 and 9070 XT and it will be more visible as more people bought these cards. Let's just hope that AMD will fix it for all cards.

[Aer]

I've been seemingly having this issue throughout owning a 7900XT

If  I'm using low GPU desktop applications or web browsing especially if I leave the PC idle for a couple of minutes I'll come back to a frozen or black screen and have to hard reset (😬hate this because of corruption). Sometimes having to reinstall the drivers.

It has been happening so often lately, so I wrote a small app Keep-Alive Render Loop. I've uploaded it to GitHub. So I'm testing it now.

https://github.com/Terrabet/KeepAliveRenderLoop/tree/releases

https://github.com/Terrabet/KeepAliveRenderLoop/releases/tag/releases

[Alto_Q]

@Aer @PlumBoniT0 

I didn't manage to launch the KARL app (something to do with my Windows restricted settings and .Net Desktop Framework) but nonetheless - I managed to fix the issue with Wallpaper Engine.

I was facing the same issues, when my idle voltage was dropping to 55 mV in idle stay, absolutely no issues or artifacts during heavy loads.

The Voltage settings and Min Freq in Adrenaling didn't do anything, so I decided to give KARL and Wallpaper Engine a try. Right now, my idle voltage is around 700 mV at idle desctop and the issue is completely gone for now.

I will do more testings but it seems that this type of issue has now at least 3 solutions - KARL, WE and Resourse Monitor. WE seems much more appealing)

Thanks a lot, you guys! I was already loosing my mind after 2 weeks of constant battling. I was so glad that Chris260999 on Reddit included your testings in his post  

[Aer]

Sounds like a good plan to run wallpaper engine, if it causes enough load to keep the GPU up.

I've not had any issues since running karl on low, so I'll keep running it.

I probably should have just left it on an old version of the framework 4.2.1.

After running WE for sometime let us know if it does cure the issue.

Also which GPU are you experiencing the issue with?

Can you fix the URL on the post? So I can check the Reddit post?

Thanks

[Alto_Q]

So far so good within 1 day of testing. It was less successfull on win 11 though, but works fine on win 10. It seems that there are a bit different algorithms for the launch process on win 11, because it can start working right off the bat after a fresh driver installation but fails to keep up after a PC restart. On win 10 there are no such issue.

I'm using PowerColor RX 7800XT Fighter. I'll try to fix the link

[Aer]

Got it and wow - quite a mega thread of issues.  I'm running windows 11 right now.

< Sapphire Pulse 7900XT

[Alto_Q]

I did some more testing, as an alternative I tried to use HWinfor 64 instead of Wallpaper Engine, it has an option to "wake up sleeping GPUs" while it is running in tray, also can be launched with Windows start.

I've installed 25.3.2 optional driver, made slight -5% max wattage restriction (to stay away from 80C on hotspot in games) and lowered Max Freq from 2637 to 2550, raised Min Freq to 1000, didn't touch default voltage slide (left on 1150mv) - the card runs good, the amount of crashes was decreased to none, but without WE or HWinfo on the background it still produces small horziontal artifacts in the browser (chrome) sometimes, I assume they potentially can cause crashes at some point. No issues with WE or HWinfo turned on, though.

Additionally, I made some changes in chrome flags (turned off everything related to WebRTC. In BIOS I switched off ReBAR and Fast Boot, just to be extra sure.

[PlumBoniT0]

Update for my crashing card at idle, I have succesfully RMAd my GPU by sending over a note within the box basically begging them to leave the card idle instead of trying to run benchmarks lol, along with the videos I had recorded of it crashing. Basically set the stage for them to realize this is a problem a bunch of us are having

The card is on it's way to me right now. Two possibilities:
1. they repaired my GPU by changing some chip on it
2. just sent me a replacement. 

But we'll know soon enough, I'll send an update this when it's back. 

@Aer , does your script target memory, clock speeds, or both? which part of the card are you pushing? 

I had also built a script for my card to keep it stable and figured out that it was the memory that needed to be kept alive specifically, not the core clocks. As long as the memory remained at +900mhz, the card would never crash. Curious if your card reacts the same as mine. I can share the script if you want, but basically just create a script that runs random memory read/writes and change the workload so that it keeps the memory clocks up. 

[Aer]

it's not loading anything to stress memory usage, It does however seem to move the vram clock speed up to various points. This wasn't intentional though.

It's literally rendering an empty canvas, with one 2D texture in DirectX (11) with 3 set loop speeds (frame rates). 100/1000/max .
(max won't literally max out your GPU but will just hit a limit)

I'm testing/running it myself on low and I'm yet to have a freeze, I've been using it for about a week and using low powered GPU tasks, web browsing, VS Code, also LM Studio where the GPU used to freeze a bit too and even left it idle. No freezing yet.
If it is Vram clocks might not even be a need for medium/high. Though if it's just core usage/voltage then might be worth keeping them.

I think I've had my GPU too long to RMA, I basically spent all the time trying to diagnose the issue with all components.  😑

Feel free to share the script or Git it. I've seen various posts about these random freezes when trying to diagnose it so it could definitely help some people.

Please do update us when it's back. I'd definitely like more information on what's going on. I've sent report upon report to AMD yet no driver has really fixed the issue.

[PlumBoniT0]

Noted, here's the script I was using, it is a python script. I have an exe version that I run as a background process using Task Scheduler, but running this from teh command line works exactly the same. 

Script: https://pastebin.com/UAm9Qe3r

These cards have a couple of memory state power buckets, so to speak. The first one (lowest power state) is basically just the memory downclocking to as low as it can, completely dynamic memory clocks. The second one makes the memory sit at 909mhz

The aim with my code was to essentially take the GPU out of the first memory power bucket using the least amount of computation / heat possible, so I could always run it and still game with no issues.

Once it's running, open up Afterburner (or your utility of choice) and make sure that it is indeed pushing the memory to 909mhz. If it isn't, increase "mem_size" on the code to 30 or 50mb. 

here's how it should look, note the minimum 900mhz memory clocks: https://imgur.com/a/7900xtx-stable-script-settings-idn8VLS

Curious to see if it works just like my card. I will make sure to update once the card is back, it might take a while as I'm on vacation right now but I'll make sure to post an update when it's back

[Alto_Q]

How do I start this app? it keeps showing me that I need .Net Desktop Runtime, but i already installed it.

[Aer]

It's on the latest .net framework 

<TargetFramework>net8.0-windows</TargetFramework>

Which that dialog should let you download and install. 

Or here from the MS site

https://dotnet.microsoft.com/en-us/download/dotnet/thank-you/runtime-desktop-8.0.15-windows-x64-inst...

Other versions:

https://dotnet.microsoft.com/en-us/download/dotnet/8.0

Let me know if it makes a difference for you.

[Alto_Q]

No, unfortunately it did nothing for me, although i tried to run it on another Windows.
I make sure to install all .NET files, but the exe keeps showing me that i need .Net to run it.

[Gamer82]

Hello friends

I have had a stuttering problem with my RX6600 with the same solution (enabling Performance Overlay). Got nowhere with customer support.. ofcourse they want to rma it but I don't think thats the the fix for this problem.

If someone wants to test a 7000 series if it has the same problem as my RX6600 with stutters on top of the stability issues?

Stutter issue RX6600 with Freesync (frametimes fin... - AMD Community

[Aer]

I'm unsure how to test this with a 7900XT I do not get any stutters as you describe in-game. I only get the PC completely freezing because the GPU/driver stops responding which requires a hard reset. It could be two separate issues.

[Gamer82]

You would have to do a 60fps limit on the game with Freesync to really get the stutters constantly.. you probably haven't played with a 60fps limit? You would have to spin the game camera around with a moderate speed to really notice it and probably have to be using a controller. 

Then you can use CapFrameX to make a capture and see if there is spikes in "Until Display Times" (under the Sync tab) when fps is almost constant 60fps. You can also see the stutter on screen or see them in the display menu where it shows the display frequency. Even though it's 60fps and the refresh rate should be around 60Hz it might jump to a lot higher number like 100 or 200Hz or drop to a much lower number when it stutters. Do not use a fps meter to detect the issue it doesn't show up in an fps meter because it is not a fps issue.

Use Freesync and Default performance tuning in Amd Software. Amd Performance Overlay need to be off for the game to stutter so when you find a place where it stutters when you spin the camera around try to enable Performance Overlay and see if it fixes the stutters. You can do another capture to verify this and maybe enable Amd Performance Overlay mid capture to verify if it fixes the stutters.

You can get Destiny 2 for free. Not as bad stutters as in Nioh for example but noticeable when spinning the character around with the controller with a moderate speed. Use 1080p for testing. If you max out the gpu it might not stutter. You need to find a place where it stutters in the game it might not stutter everywhere depends on the game.

[PlumBoniT0]

Your case is interesting, and the first thing that comes to mind is that...since you're running such a low resolution, and your GPU is basically chilling running it, the power budget that the card has is extremely high, and the card is trying to use that power budget to push the memory clocks outside of spec, which causes instability and stuttering. Obviously just a theory, and it could be infinitely more complex than that. GPUs are extremely complex devices.

If you're stuttering even with normal games under normal resolutions, and you're sure it's not that you're CPU bound / single-threaded performance limited, then your GPU might just be like that. 

One thing I can assure you though, after months of talking to various people within the GPU industry, that know these GPUs inside out. AMD will not release a fix for your singular GPU, and probably not for ours either.

The kinds of things that get fixes by AMD are generally widespread problems, not one-off scenarios, especially not from old architectures like RDNA2 and much less while running custom resolutions. It's been a **bleep** good while since RDNA2 was mentioned in the driver notes..

I would totally suggest just using the fix you know works (maybe try the script I just posted on another reply since your problem seems to also be vram related), or RMAing your card if you get tired of stuff not working as it should. 

It's the best way to go about it, because AMD (nor NVIDIA!) push updates for this kind of oddball stuff. It's not your fault that the card is not running as expected. I've found that the sooner you get familiar with that, the quicker you'll fix your issue. Either use the fix you know works, or RMA the card, or sell it and get something else. Good luck

[Gamer82]

Well the thing is I don't think it's just me. Nobody else (that I know of) has tested this because they probably feel they would have noticed. I don't think people would notice this in day to day use. I have had this card for over 2 years and just now noticed how I could get it to stutter all the time basically.

Spoiler

I have the problem actually in 1080p high settings in many games so it's not a small resolution issue. You just have to cap the fps to 60 or if the game has a fixed 60fps cap (which many games do) and if the gpu usage is too low it starts to stutter. In many games the gpu usage changes a lot and sometimes it stutters even if you are not limiting the fps. Some games doing 1440p res for example could kinda fix it and not all games seem to stutter.

I have had so many issues with RMA. When a problem feels software related it usually is. Last time I RMA'd something (that felt like a software issue) Samsung tried to fix my TV's picture quality issue by changing basically everything in that TV and it was still the same. After that they just wrote it off as a "feature" and I just sold it to someone else.