Weird GPU performance issue

[Graf Zahl]

Recently I noticed with my graphics card that when I start a game it runs at a good performance - but after 30-60 seconds it always seems to power down a bit and performance drops noticably. So far I haven't found anything obvious - no dust, no loose cables, no anything. From the outside it all looks fine.

Does by any change somebody here have a clue what else I could check? The card in question is a Geforce 1060, with the latest driver as of this writing.

[Rachael]

You will probably want to install some sort of monitor that gives you a graph charting your performance with temps, fan speed, and clock speed over time.

One such program is MSI afterburner - but keep in mind that this program is meant for actually adjusting such values. It should be possible to set the program up so that it only observes it, that should give you some idea what in the heck is going on there. If the driver is controlling these values, then you will not be able to adjust them in the program.

You do not need any sort of MSI hardware to use this program.

If someone has an idea for a program that can also create graphs like this one, except without actually having the sliding adjustments - that would definitely be a superior option to this one for your case.

---

Anyway - most important thing to monitor is your GPU temperature. If it seems to be going above 70*C you may want to lower the clock speed and see if you can get it within the 60's range - if it takes a lot of adjusting to get it down, then it's probably time to redo your thermal paste.

[Graf Zahl]

I'm not sure what to make of this. MSI Afterburner reports a temperature of 24°C in idle mode and 28°C when the performance drop occurs. But these values seem extremely low.

[Rachael]

Yes - that is very VERY low. Your GPU should be able to run up to 50C - possibly even 60C - before the fan even kicks on.

It makes me wonder if the driver is doing something it's not supposed to be doing. Or if there's another process on your system that's forcing your GPU to run at a reduced clock rate - or if it's even doing anything at all. 28C is shockingly close to ambient temperature.

[Rachael]

One thing to try is to run GZDoom without any frame rate limiters enabled. See if you can get a super high frame rate. The important thing here is to see if the fan turns on - one hypothesis that I have here is that the driver is detecting that the fan is not working, so it is refusing to give the GPU any sort of workout at all for fear of damaging it.

[Graf Zahl]

Thanks for the tip. The fan does indeed not spin up when running a high frame rate. With my old card it never took more than a minute for that to happen, but I left it running for several minutes now and no increase of the noise level. The fan still operates at idle speed.

Now, what now? I'm not a technician who can analyze this. What I can say is that the fan works. If I force it to run faster it will do so. The thing I do not trust is the temperature display, right now it says 23°C, and the card is set to increase fan speed when exceeding 30°C, but even after playing for several minutes the value is still below.

10 minutes later. The temperature is slowly going up and the fan getting louder, so this isn't it I guess...

[Rachael]

Okay, so the issue is not as serious as I suspected. So it's probably just a misbehaving application.

One of the worst ones is of NVidia's own writing - GeForce experience. It doesn't muck around with GZDoom settings, but it sure likes to fuck up my settings of any game it recognizes and I hate that. If I do a driver update and I notice that it's there, it's the first thing I uninstall. I don't have an NVidia Shield device to stream to, so I don't even need that atrocity anyway. (Hypothesis: Possibly, it sees that GZDoom is not very demanding on the card so maybe it's trying to run the card in a power saving mode so that you still get what it considers high performance but low heat output)

Outside of that, I'm not able to think of anything at the moment that'll cause it to behave that way.

[Graf Zahl]

I don't think it's some app misidentification. I get it on GZDoom's Vulkan branch version with both backends and the EXE is called gzdoomrd.exe (RelWithDebug) so not even the name matches.
This was one of the first things I tried, but it made no difference and the fact that it happens with Vulkan as well hints at something different.

[Rachael]

Well - I'm sorry, but I am at the limits of my knowledge figuring out what could be causing that. If I knew what was running on the system I may have some clue, but at this point I doubt even that much because I have never seen anything lower my GPU intentionally without my interaction unless I close the lid on my laptop or ran the temps up too high.

But as I said - GeForce Experience has always been a source of fuckery for me, and god knows what new "features" have appeared in that monstrosity ever since I instituted the policy that it's expressly forbidden from ever appearing on my computer - or if it does, that it's instantly nuked out of sight on the moment's notice.

[Graf Zahl]

I do not have that installed, so it can't be the cause.

[Rachael]

Only other idea I have is possibly a programming error that's somehow locking up the CPU every cycle progressively more and more and preventing the GPU from doing any work. (Similar to a memory leak?) I do not know if that's the case - but it seems feasible. If that's the case, it could come from the driver, or Windows, or GZDoom, or possibly anything in between.

If that is the case, then GZDoom will always start with a consistently high GPU processing rate before dropping, without you having to wait between launches to observe the high start.

[drfrag]

I think it's a driver malfunction. You could try changing power management mode in nvidia control panel from adaptive to maximum performance as a workaround.

[Graf Zahl]

I am already on that setting. It was the first thing I changed when I noticed this behavior. No driver update made any change to it in all the time.

[SHayden]

Are you sure it's not a hardware issue? I'm no expert by no means but last time something like that happened to me it was the PSU

[Graf Zahl]

How can I test that?