Will abnormal GPU usage hurt my GPU?

[Atticus_Maytrap]

Hi there,

yes, the question would seem to answer itself, but let me elaborate.

For the past few month, ive been seeing odd spikes in GPU clock + usage with desktop programs that have never needed so much juice in the past, including:
*Autocad
*Revit
*Adobe Premier Pro
*Adobe Indesign (but only when typing text)

Basically, the GPU suddenly seems to treat using these programs as though they were games. If i just run the mouse over an empty Revit file, the GPU clock shoots up to load levels and the usage spikes to 70-80%.
Strangest things is that in other just as 3D intense programs like 3DS Max or Sketchup, the GPU clock+usage behaves as it should, never leaving the base clock and only using about 10-15% tops.

I'm still running tests on different things to see where this problem is coming from - Nvidia drivers, Windows or BIOS are my lead suspects. I've rolled back drivers and the problems gone away (very telling), however it
reappeared out of nowhere after 2+ weeks of normal use on the most recent driver - HOWEVer, while it is certainly abnormal behaviour, the computer still runs great. GPU heat never gets out of control, game still run
fine and the affected programs mentioned above still work as expected.

The question is: is it okay to work with the machine while the GPU is bouncing up and down to such extremes like that? The clock remains steady at load levels (like it would for games) while using these programs, but the usage is up and down like
a schizophrenic rollercoaster.

 

[Chomiq]

Hi there,

yes, the question would seem to answer itself, but let me elaborate.

For the past few month, ive been seeing odd spikes in GPU clock + usage with desktop programs that have never needed so much juice in the past, including:
*Autocad
*Revit
*Adobe Premier Pro
*Adobe Indesign (but only when type text)

Basically, the GPU suddenly seems to treat using these programs as though they were games. If i just run the mouse over an empty Revit file, the GPU clock shoots up to load levels and the usage spikes to 70-80%.
Strangest things is that in other just as 3D intense programs like 3DS Max or Sketchup, the GPU clock+usage behaves as it should, never leaving the base clock and only using about 10-15% tops.

I'm still running tests on different things to see where this problem is coming from - Nvidia drivers, Windows or BIOS are my lead suspects. I've rolled back drivers and the problems gone away (very telling), however it
reappeared out of nowhere after 2+ weeks of normal use on the most recent driver - HOWEVer, while it is certainly abnormal behaviour, the computer still runs great. GPU heat never gets out of control, game still run
fine and the affected programs mentioned above still work as expected.

The question is: is it okay to work with the machine while the GPU is bouncing up and down to such extremes like that? The clock remains steady at load levels (like it would for games) while using these programs, but the usage is up and down like
a schizophrenic rollercoaster.

View attachment 232815

GPU temps are flat. No reason to worry.

 

[Atticus_Maytrap]

GPU temps are flat. No reason to worry.

awesome, this is what i assumed but the abnormal behaviour of the GPU still bugs me

any idea what could be causing it?

that screenshot is what happens when i switched over from my browser to an empty (no 3D geometry modelled at all) Revit file and just zoomed in and out / panned around. Spikes in usage, BUS and core clock. For this to happen in a 3D modelling program might not seem unusual but it is def. not the norm. As i said, working in other 3D programs (3DS Max, Sketchup) the usage never gets past 20% and the core clock stays at base levels. In the past Revit and Autocad behaved the same way, but now all of a sudden theyre seemingly very resource greedy for no reason at all.

That the spikes also occur in other prgrams like InDesign is really bizarre, right? And even then, only when im typing in text??

 

[eidairaman1]

Stop worrying about it

 

[Atticus_Maytrap]

Stop worrying about it

the machine runs fine so sure, no worries - but it is weird shit though, right?

 

[TheoneandonlyMrK]

the machine runs fine so sure, no worries - but it is weird shit though, right?

It's possible something you aren't considering is using the GPU, for example Corsair icue uses open GL I think but more generally the GPU to process lighting effects, odd but true.

 

[Atticus_Maytrap]

I've not got any lighting effects, no time or need for any of that.
the Task Manager says its is specifically these programs that are using up 3D resources when it happens, but then i dont even have to really "use" them, just run the mouse over them and bam, up goes the GPU clock.
That it happens in some 3D programs but not others, and then also in some other programs that are very much not 3D makes it a real head scratcher.

I typically use 3 monitors for work, and turning off monitors 2 and 3 when i game.
Now when i turn the other monitors back on, this also causes the GPU clock to spike up from base to load levels. Give it a few minutes and the clock eventually stabilises.
This never used to happen.

I am currently on driver 511.23, i did a clean install of it on Saturday and it made no difference. The sequence of events for me has been:



Driver 466.47, everything running as it should be.

Drivers 466.77-471.12, all having the upclocking bug with regard to certain programs.

Roll backed to 466.47, everything running normal again.

Rolled forward to driver 496.13 on release, everything still running normal (*see below)

skipped over the next few drivers due to negative user reviews.

Rolled forward to driver 497.29, everything still running normal.

after a few weeks, encountered some intermittent poor performance with a few games (GPU usage dropping to zero and causing stuttering).

Reinstalled the same driver after using DDU in safe mode to remove it.

Upclocking bug reappears after fresh install of driver 497.29.

Rolled back to 496.13, upclocking bug still there (where it wasnt before)

Roll forward over to studio driver 511.09, everything running normal

work for a few hours, load and create some real time renders with Enscape, GPU clock behaving exactly as it should - clocking up for the renders, clocking down again after once the render window is closed.

Put machine into sleep mode to go for a walk, come back and the upclocking bug is there again on studio driver 511.09

Rolled forward with a fresh install of 511.23, upclocking bug still there.



* It's also worth pointing out that while on driver 496.13 the upclocking wasnt there (for a time at least) with Revit, but WAS there with other programs like Autocad or Adobe Premier Pro

im currently using Revit, Autocad and Adobe indesign on my office machine and the GPU clock and usage are behaving as expected.
So strange.....

 

[Toothless]

Nothing strange besides your obsession after being told it's fine. Let the machine do its thing.

 

[Atticus_Maytrap]

Nothing strange besides your obsession after being told it's fine. Let the machine do its thing.

what can i say man, if i see the machine doing something it didn't do in the past im eager to figure out what's what.
Last thing i want is to ignore it and have it fuck up my GPU later on.
Buying a new one now? Forget about it.......

 

[Udyr]

what can i say man, if i see the machine doing something it didn't do in the past im eager to see what's what.
Last thing i want is to ignore it and have it fuck up my GPU later on.
Buying a new one now? Forget about it.......

Then reinstall your OS and see if everything goes back to what it was in the "past".

 

[Atticus_Maytrap]

Then reinstall your OS and see if everything goes back to what it was in the "past".

it's on the to-try list - right at the bottom of course, but it's always there

 

[Tetras]

The question is: is it okay to work with the machine while the GPU is bouncing up and down to such extremes like that? The clock remains steady at load levels (like it would for games) while using these programs, but the usage is up and down like
a schizophrenic rollercoaster.

View attachment 232815

Buildzoid did a video on this sort of stuff, I think it was this one:

So.., yeah, it could be a problem hypothetically, but the voltages and temperatures seem to be barely moving and they're the biggest issue. If you want to be 'safe', then just continue to avoid the bug.

 

[eidairaman1]

It's possible something you aren't considering is using the GPU, for example Corsair icue uses open GL I think but more generally the GPU to process lighting effects, odd but true.

Could be a trojan/ghost miner

 

[Vayra86]

awesome, this is what i assumed but the abnormal behaviour of the GPU still bugs me

any idea what could be causing it?

View attachment 232816

that screenshot is what happens when i switched over from my browser to an empty (no 3D geometry modelled at all) Revit file and just zoomed in and out / panned around. Spikes in usage, BUS and core clock. For this to happen in a 3D modelling program might not seem unusual but it is def. not the norm. As i said, working in other 3D programs (3DS Max, Sketchup) the usage never gets past 20% and the core clock stays at base levels. In the past Revit and Autocad behaved the same way, but now all of a sudden theyre seemingly very resource greedy for no reason at all.

That the spikes also occur in other prgrams like InDesign is really bizarre, right? And even then, only when im typing in text??

What's more strange to me than it 'spiking', which just means 'usage' and your GPU clocks to boost clocks just fine and backs down after...

You seem to be on a constant base clock, the GPU never enters a lower power state. I think that's the angle you should look into, not so much the spikes but the fact it won't clock down. It really should, unless an application just forces it into 3D clocks all the time. And if it does, its not strange an increase in load will cause a 'spike'. I've seen it in some OpenGL stuff too, being very keen to maintain base clock and not allowing dynamic clocking all that well.

Either way, GPUs are designed to have wild usage fluctuations all the time, similar to CPUs. Transistors want to switch That's what its doing.

 

[Atticus_Maytrap]

What's more strange to me than it 'spiking', which just means 'usage' and your GPU clocks to boost clocks just fine and backs down after...

You seem to be on a constant base clock, the GPU never enters a lower power state. I think that's the angle you should look into, not so much the spikes but the fact it won't clock down. It really should, unless an application just forces it into 3D clocks all the time. And if it does, its not strange an increase in load will cause a 'spike'. I've seen it in some OpenGL stuff too, being very keen to maintain base clock and not allowing dynamic clocking all that well.

Either way, GPUs are designed to have wild usage fluctuations all the time, similar to CPUs. Transistors want to switch That's what its doing.

the constant base clock is just because i have multiple screens with different refresh rates. When they're turned on, the card always just sits at base, it's just the way it is.
When i turn them off it clocks down to a lower power state just fine.

And yes, i have tested the issue i have here with only my main monitor on. It'll go from idle at 130+MHz and slam all the way up to 1900MHz+ just by running the mouse over an offending app - apps that i know for a cold hard observed fact do not need the card to be at load levels like that.
For example, aside from Revit & Autocad, if i just run the mouse over the graphs in Afterburner, the card clocks up from base to 1900MHz. Another one is Adobe Indesign. I can work in a document just fine aligning pictures and objects etc. but the minute i type text in a text box, up slams the GPU clock.
Meanwhile i can model crazy complicated mesh objects in 3DS Max or Blender while streaming video on one of my extra monitors, and the GPU clock never leaves base and usage stays down around the 20-30% mark, like it always has.

And of course you're 100% correct about GPU fluctuations. What im here seeing here though - for example with Revit, that's what was being used in the screenshot from from original post - is fluctuations that are scaled up to 2-3x from what i know to be normal.
Im using Revit right now on my office machine, card clock is at base levels, usage between 10-20% with video streaming on an extra monitor.
I use it at home on my machine there, card goes to load levels and usage up between 70-80% just by running the mouse over an empty file with no geometry modelled at all or other programs running.

This is abnormal behaviour. This is why i'm here (and on Tom's Hardware, and Reddit, and the Nvidia Forums, and the MSI forums......) bothering good people like yourselves.
If it were normal, i promise you i would have seen it before since ive always kept an eye on things like clocks & temp. & usage while im working.
But it started after an Nvidia driver update, went away after rolling back a driver, and now has returned and wont go away again no matter what i do.

......but like the man said, temps are fine, performance is also fine, so why worry?
Short of reinstalling windows i can certainly live with it.
I guess it's just my need to know, you know?

 

[Mussels]

No, that's not abnormal. It's the observer effect in a high demand situation.


You're running the card in a high stress environment with the multiple high refresh rate monitors, monitoring it (yes, this adds to the load) and then running into situations the driver thinks requires greater performance, whether you think it does or not.


Many GPU's over the years have had corruption and crashing issues with multiple high bandwidth displays, and the solution was always to raise the default clocks - it'd get better the next generation, until higher bandwidth and higher refresh rate displays came along...

Those are IDLE multi monitor levels. It varies wildly, and the only thing that needs to have changed to throw your experience out the window is that you're running a different GPU, different monitors, or different monitor technologies (add in Gsync, freesync, or changing between HDMI and displayport connections) and suddenly behaviour is wildly different
Windows updates, OS change, hardware scheduling, driver updates... so many variables here. Even changes to how something like afterburner gets readings from the GPU could result in the card wanting to spike higher to a demand it thinks is coming.

 

[Vayra86]

the constant base clock is just because i have multiple screens with different refresh rates. When they're turned on, the card always just sits at base, it's just the way it is.
When i turn them off it clocks down to a lower power state just fine.

Short of reinstalling windows i can certainly live with it.
I guess it's just my need to know, you know?

Great. And yes, I understand your question. The best we can get to I think is educated guesses.

From (recent) history:

https://www.nvidia.com/en-us/geforce/forums/geforce-graphics-cards/5/233230/very-high-idle-clocks-on-gtx-1080-with-multiple-di/

https://www.nvidia.com/en-us/geforce/forums/game-ready-drivers/13/224939/144hz-monitor-high-idle-clocks/

This aligns with your constant high base clock as you've also explained correctly. But I think this is also the whole answer after all:
Consider what you're stacking on top:
- you have multiple screens with different refresh rates.
- you're running 2D (desktop) loads alongside (windowed) 3D loads, and while your application window is in focus, the GPU still can't allocate all performance to it - it is still running a much larger desktop with a low load.
- We've seen Nvidia (and AMD) tweaking the crossover point between idle clocks and base clock because of complaints about responsiveness. You can strike a different balance: less responsive, more power saving for example. Note the NV inspector comments up here.

I've also experienced a similar thing moving to high refresh back in the day on different GPUs. And similarly, with GPUs going for 3D clocks at moments you'd not always consider logical. Indeed after driver updates, too. We've also seen differences in power management between Nvidia driver updates, mostly pre-and post-Maxwell release. GPUs and monitor resolutions have been getting stronger in tandem lately. Its very likely changes had to be made to favor more, newer combinations of hardware to hit optimal performance.

No, that's not abnormal. It's the observer effect in a high demand situation.


You're running the card in a high stress environment with the multiple high refresh rate monitors, monitoring it (yes, this adds to the load) and then running into situations the driver thinks requires greater performance, whether you think it does or not.


Many GPU's over the years have had corruption and crashing issues with multiple high bandwidth displays, and the solution was always to raise the default clocks - it'd get better the next generation, until higher bandwidth and higher refresh rate displays came along...

View attachment 233370
Those are IDLE multi monitor levels. It varies wildly, and the only thing that needs to have changed to throw your experience out the window is that you're running a different GPU, different monitors, or different monitor technologies (add in Gsync, freesync, or changing between HDMI and displayport connections) and suddenly behaviour is wildly different
Windows updates, OS change, hardware scheduling, driver updates... so many variables here. Even changes to how something like afterburner gets readings from the GPU could result in the card wanting to spike higher to a demand it thinks is coming.

Haha minds thinking alike AND at similar times. Dayum

 

[Atticus_Maytrap]

No, that's not abnormal. It's the observer effect in a high demand situation.


You're running the card in a high stress environment with the multiple high refresh rate monitors, monitoring it (yes, this adds to the load) and then running into situations the driver thinks requires greater performance, whether you think it does or not.


Many GPU's over the years have had corruption and crashing issues with multiple high bandwidth displays, and the solution was always to raise the default clocks - it'd get better the next generation, until higher bandwidth and higher refresh rate displays came along...

View attachment 233370
Those are IDLE multi monitor levels. It varies wildly, and the only thing that needs to have changed to throw your experience out the window is that you're running a different GPU, different monitors, or different monitor technologies (add in Gsync, freesync, or changing between HDMI and displayport connections) and suddenly behaviour is wildly different
Windows updates, OS change, hardware scheduling, driver updates... so many variables here. Even changes to how something like afterburner gets readings from the GPU could result in the card wanting to spike higher to a demand it thinks is coming.

Ok, but again, why would it consider typing text in Indesign high demand - or just running the mouse over the graphs in MSI AB - , but modelling furniture with NURBS surfaces in 3Ds Max not high demand?

......and i really cant stress this point enough, it did not use to do this.
These specific programs - some of them completely non-3D programs - that are suddenly producing a high demand load reaction from the GPU NEVER used to do this.
It is new - if unproblematic - behaviour, that started after a driver update, that was then solved by rolling back the driver, but now that doesn't work either.

 

[Vayra86]

It is new - if unproblematic - behaviour, that started after a driver update, that was then solved by rolling back the driver, but now that doesn't work either.

This is interesting, did you clean install the rolled back driver? And did you then clean install the new one again?

Also, did you reinstall the applications where it clocked higher post driver updates? Might very well be that the highest demanded load is logged somewhere and referred to. GPU drivers have stuff in place to predict loads, again to be more responsive, also there are things like driver cache that saves stuff per application.

 

[Atticus_Maytrap]

Great. And yes, I understand your question. The best we can get to I think is educated guesses.

From (recent) history:

https://www.nvidia.com/en-us/geforce/forums/geforce-graphics-cards/5/233230/very-high-idle-clocks-on-gtx-1080-with-multiple-di/

https://www.nvidia.com/en-us/geforce/forums/game-ready-drivers/13/224939/144hz-monitor-high-idle-clocks/

This aligns with your constant high base clock as you've also explained correctly. But I think this is also the whole answer after all:
Consider what you're stacking on top:
- you have multiple screens with different refresh rates.
- you're running 2D (desktop) loads alongside (windowed) 3D loads, and while your application window is in focus, the GPU still can't allocate all performance to it - it is still running a much larger desktop with a low load.
- We've seen Nvidia (and AMD) tweaking the crossover point between idle clocks and base clock because of complaints about responsiveness. You can strike a different balance: less responsive, more power saving for example. Note the NV inspector comments up here.

I've also experienced a similar thing moving to high refresh back in the day on different GPUs. And similarly, with GPUs going for 3D clocks at moments you'd not always consider logical. Indeed after driver updates, too. We've also seen differences in power management between Nvidia driver updates, mostly pre-and post-Maxwell release. GPUs and monitor resolutions have been getting stronger in tandem lately. Its very likely changes had to be made to favor more, newer combinations of hardware to hit optimal performance.



Haha minds thinking alike AND at similar times. Dayum

i see what you're saying, and i don't want to be a dick wasting your time with what appears to be very much a non-problem (performance and temps, no complaints), but this reaction from the GPU happens even when i only have my main screen running and is programs specific - programs that i know never used to do this.

Another offending program is Driver Easy. I dont have it installed on my machine permanently, i load it on once every 6 months or so just to check on things, but just opening this program - before any scanning takes place - with NOTHING else open causes the clock to crank up from idle to gaming levels. And again, i cant stress this point enough, it did not use to do this.

Meanwhile, proper 3D apps like 3DS Max - on both single and multiple monitor situations - do NOT produce the same effect.

That is weird, right?

This is interesting, did you clean install the rolled back driver? And did you then clean install the new one again?

Also, did you reinstall the applications where it clocked higher post driver updates? Might very well be that the highest demanded load is logged somewhere and referred to. GPU drivers have stuff in place to predict loads, again to be more responsive, also there are things like driver cache that saves stuff per application.

i did, back when it first started happening i rolled back to driver 466.47 and the odd GPU reactions disappeared.
Every driver after that up until the newer windows 10+11 drivers had the reactions - always with the same programs.

Drivers 496.13 and then 497.29 also worked fine. Then i had some stuttering in a game (GOTG, lots of fun) and tried DDU'ing and then a clean install of the same driver (497.29), and the reactions started again.

If you can be bothered (i fully understand if you cant) there's an earlier post from me up near the start that shows a basic timeline of what i've been seeing.
It's a god damn puzzle man!

.....and yes, tried reinstalling the specific programs that are producing the reaction, didnt change anything

 

[Vayra86]

I've looked at your posts, and I still stand by the earlier explanation. It fits.

You even sort of confirmed it yourselves, as you've seen stuttery behaviour in applications. The question is now: does your current, up to date driver and Windows installation etc. still produce stutter?

If it does not, that only confirms the theory further: GPU knows what you're running and wants to deliver sufficient grunt to do it stutter free.

 

[Atticus_Maytrap]

I've looked at your posts, and I still stand by the earlier explanation. It fits.

You even sort of confirmed it yourselves, as you've seen stuttery behaviour in applications. The question is now: does your current, up to date driver and Windows installation etc. still produce stutter?

If it does not, that only confirms the theory further: GPU knows what you're running and wants to deliver sufficient grunt to do it stutter free.

but the stuttering was in just one game? I never had stutters in anything else?
I could see the activity of the stuttering in the afterburner graph after i quit out of the game, the usage was dropping from 95%+ down to virtually null and then back up again.
at first the fix for this was just to restart the machine, game stutter just vanished. When restarts stopped working, i tried the fresh install of the driver, and then oddball program specific GPU reactions started up again.

admittedly ive not played that particular game since these GPU reactions started again, but i have played others.
Performance is perhaps better in BF2042 but then that has had updates recently. The average temp of the card while playing is also lower. It used to be around the 65-70'C mark, but last night it didnt even get past 60'C.

Just to recap and see if i understand you correctly, you're saying that the card is compensating for previous under-performance in a particular app by applying more power, and this effect is then spilling out into other apps? Apps where ive never had performance issues at all?

 

[Frick]

Have you looked at what is actually using the GPU?

 

[Atticus_Maytrap]

Have you looked at what is actually using the GPU?

yes, Task Manager always states plainly that its the programs in question that are producing the reaction from the GPU

 

[Vayra86]

but the stuttering was in just one game? I never had stutters in anything else?
I could see the activity of the stuttering in the afterburner graph after i quit out of the game, the usage was dropping from 95%+ down to virtually null and then back up again.
at first the fix for this was just to restart the machine, game stutter just vanished. When restarts stopped working, i tried the fresh install of the driver, and then oddball program specific GPU reactions started up again.

admittedly ive not played that particular game since these GPU reactions started again, but i have played others.
Performance is perhaps better in BF2042 but then that has had updates recently. The average temp of the card while playing is also lower. It used to be around the 65-70'C mark, but last night it didnt even get past 60'C.

Just to recap and see if i understand you correctly, you're saying that the card is compensating for previous under-performance in a particular app by applying more power, and this effect is then spilling out into other apps? Apps where ive never had performance issues at all?

Not so much compensating, or spilling over to other apps, but rather, the initial runs were choppy and post-driver update the card manages the loads better, boosting earlier, to make sure it never gets to those dips in performance. The card has numerous ways to predict what clock it should produce but the simple gist of it is: it always wants to clock high enough to provide smooth performance, influenced by its power plan, if applicable. And drivers have been seen to change the metrics the GPU uses to determine that. Cards, for example, wouldn't always recognize 3D loads properly, especially as power saving features got introduced. I think that's what you're seeing and what caused stutter - GPU load zero is either game logic, system interrupt, or a driver bug. The whole affair regarding multi screen and high refresh monitors is of a similar nature: using the correct clock profile for the load it gets.

Note, that lower temps can also indicate the presented loads are handled better, or more efficiently, though a 5C gap kinda tells me you're just using different settings somewhere.

Other than that, I'm just working with whatever you're giving me. Its a simple fact that the GPU is working properly, boosting properly, and responds properly to a presented mix of loads, which is inherent to your setup at all times, so the only sensible idea is to be looking at:

- combinations (at moments in time) of updates/versions of software, drivers, BIOS, Windows. I find it hard to believe you've had a perfect testing bed to compare both situations and we've also seen Windows updates bork GPUs or fix them again, or requiring further TLC from GPU manufacturer.
- known changes in the driver regarding power/boost management