Community
cancel
Showing results for 
Search instead for 
Did you mean: 
Cody
New Contributor I
52,582 Views

BUG: dwm.exe uses memory leakage with Intel HD Graphics 630

Hi,

dwm.exe (Desktop-Window Manager) uses high memory with Intel Intel HD Graphics 630. This problem has been reported before:

https://answers.microsoft.com/en-us/windows/forum/all/desktop-windows-manager-dwm-memory-leakage-in/...

 

https://www.reddit.com/r/windows/comments/iqttw5/windows_10_dwmexe_high_ram_usage/g4umzyn/?utm_sourc...

 

Here's a screenshot of my dwm ram usage after 2 hours (1.6 GB):
Image1.jpg

 

Please, this bug is already known for 3-4 months, and I do not want to reinstall a 2019 driver...

230 Replies
Cody
New Contributor I
5,466 Views

Thanks, I also rolled back now to 26.20.100.7985 and can confirm the bug is gone in this version, too. I also experimented a bit with nvidia and e.g., disabling the drivers had no affect on the bug.

 

So for now I can state the following: something between 26.20.100.7985 and 27.20.100.8853 happened that creates a memory leak when using the fast startup (shutdown). On restart (fresh boot), this bug does not appear in 27.20.100.8853. @vitalyplatoff can you try whether the leak is gone on restart for you, too?

vitalyplatoff
Novice
5,463 Views

@Cody Ok, i will try to check this

RonaldM_Intel
Moderator
5,360 Views

Hello everyone,

I also agree with you that this looks to be a bug, though I'm not sure - yet - if it is OS or graphics driver, obviously that would be up to Microsoft's or Intel's debug teams to figure out.

Before this report can reach Intel's debug team we definitely need to come up with a reproducible scenario... from personal experience I can tell you how frustrating it is when I get to see a bug report bouncing back to me because the debug team can't reproduce it on their systems.

I am trying to reproduce the issue as well on my NUC7i5BNK using Windows 10 version 2004 (build 19041.610) and drivers 27.20.100.8681 and 27.20.100.8853.

My test methodology so far is as follows:

  1. Watch some series episodes on Netflix (average of 1 to 1.5 hours of streaming).
  2. Open 5 or 6 additional tabs in Chrome.
  3. Let the computer on for 2-3 hours.
  4. Hit Hibernate and wait for 20 minutes to resume.
  5. Observe at all times Task Manager for dwm.exe and if no change, I do a Shut Down using start menu.

As for fast-startup, here's my current configuration:

config.PNG

Please let me know if I am missing something or I should be doing something differently.

Best Regards,

Ronald M.

Cody
New Contributor I
5,348 Views

@RonaldM_Intel 

Hi,

 

much thanks for trying to reproduce this bug. I observed the behaviour a bit more, and watching videos is not necessary. I recorded a vlc desktop video (length 3 minutes maybe), you see the dwm needing 211 MB. Alone with moving around the window I reach 300 MB ram after 1.5 Minutes aready (dwm is for graphical effects for windows, so moving around a window is enough to see whether this bug appears or not). You see the ram use accumulating.

Then you see me restarting the dwm process and the bug is gone (dwm goes maybe up a bit while moving the window but then it goes down again). The driver is 27.20.100.8853

https://streamable.com/knsc8j

 

Here's again what "fixes" this bug:

- roll back driver to 26.20.100.7985

- restart the comouter instead using fast startup through shut-down

- restart the dwm.exe in task-manager

 

RonaldM_Intel
Moderator
5,342 Views

Thanks for the video and your cooperation! 

I tried doing the same (except I recorded the video using OBS) but moved the windows around for about 1.5 minutes and still can't see DWM utilization increasing nor doing it 20 minutes afterwards (it average at around 90MB).

The recording of this test is here.

I'm not giving up yet - far from it actually - I'll keep my computer ON overnight and use it as normal while keeping an eye for any memory leak. I'll report my results back tomorrow.

Just in case I still can't see the issue on my computer... let's start thinking outside the box:

  • What is the screen resolution/refresh rate you are using?
  • Do you have more than one active Display? if so, how many? what's their respective resolution/refresh rate?
  • Is your computer using Single-channel memory or Dual-channel?

For fairness sake, my NUC is using a clean OS installation (I installed the OS a couple of days ago) and I did install all available updates via Windows Update and the only apps I have installed are Steam, OBS Studio and the Unigine Heaven benchmark.

Regards,

Ronald M.

Cody
New Contributor I
5,319 Views

Much thanks for your caring. It's indeed strange that a percentage of users have this bug but I think the majoritiy doesn't have it (otherwise it would have been reported much earlier and more frequently). I updated windows to the preview version (also clean install, and the bug was still present). There might be other system ressources which might trigger this bug (like I suspected nvidia, but disabling that card doesn't make this bug disappear). I may try to play around with the task scheduler and disable some stuff there).

 

Regarding your questions (trying to translate):

1:
Desktopresolution:1920x1080

active signal resolution 1920x1080

rate of actualization 60.020 Hz

bitdepth: 8-bit

color format: RGB

colorroom: SDR

 

2:

only the notebook display (hp pavilion 17-ab402ng)

 

3.

I think you are referring to the ram slots used, I got this notebook with 1x8GB ram and upgraded the second slot with another 8 GB, so I have 2x8GB=16GB Ram

vitalyplatoff
Novice
5,297 Views

@RonaldM_Intel , @AndrewG_Intel Hello!

Maybe it will be useful for debug team - I just power on my laptop, open spotify and PowerPoint and have almost 1000 mb ram usage. Send you screens of that. 

 

RonaldM_Intel
Moderator
5,285 Views

Hi @vitalyplatoff @Cody 

As I mentioned yesterday, I left my NUC on overnight and also did once in a while a Sleep and Hibernate cycles to see if there was any change in dwm.exe memory. It behave normally (never going above 150MB and coming down to less than 80MB afterwards).

I resized windows and moved around for hours to no avail. I also used the new input from @vitalyplatoff and launched Spotify, PowerPoint, Word and still no reproducible scenario. Just to explore a bit more I also tested by doing Win + Tab (several times) to launch the Desktop manager and this is the only instance I saw dwm.exe go above 200MB mark (only to go back to normal afterwards). See the recording I made here

I am also testing this issue on my Thinkpad T580 laptop but once again, it works fine on my side.

Can any of you test by doing a clean OS installation and verify if the issue occurs in that instance? It's a long shot, but I am suspecting an OS update might have introduced an update to dwm.exe that triggers the issue when paired with our new drivers... at least that might explain why not everyone sees the problem (perhaps they haven't received said update).

Best Regards,

Ronald M.

Cody
New Contributor I
5,272 Views

@RonaldM_Intel 

I made a clean installation back to 20H2 (didn't want to keep the preview version anyway).

 

Unfortunately, it did not solve it. The only way I have now is to roll back the driver to 26.20.100.7985 and hope you are able to find out what may cause this - in case you can somehow reproduce...

Cody
New Contributor I
5,255 Views

Just saying that on 26.20.100.7985 the bug is not present. @RonaldM_Intel , would it help for your debug team if I test all drivers from then:

 

to see with which update the bug first happened? Maybe you can find out in the update logs whether something was tinkered in the hibernate.sys or stuff like this?

 

RonaldM_Intel
Moderator
5,229 Views

Thanks for offering your help @Cody although I think driver wise we already know the most important data:

Issue appears on:

  • 100.8681
  • 100.8853 (latest to date)

Issue does not appear on:

  • 100.6290
  • 100.7990
  • 100.7985

However, the current challenge is finding a reproducible scenario so I can report this to debug (with confidence they will be able to see the issue in their lab). In both systems I have tested this the bug is not appearing, which would suggest there is another variable that is triggering the bug.

Edit: I just thought of another variable that might explain why the issue is only appearing on some computers... and that would be the BIOS. Have any of you updated the BIOS on your computer recently? I know it would seem hard to explain why we appear to be moving away from the graphics driver as the sole culprit here (especially since older versions work fine) but there are instances in which it turns out the new driver is actually doing the right thing and the issue is introduced somewhere else (e.g. OS update, BIOS) with direct conflict with the new drivers .

I'll leave this discussion open for other users to chime in, but at the same time I would recommend you all report this to Microsoft so they can do the 1st level debug. Rest assured that if Microsoft confirms the bug is triggered due to an issue with Intel driver's code they will work with us directly to address it.

Best Regards,

Ronald M.

Cody
New Contributor I
5,224 Views

Ok, I'll stick then to 26.20.100.7985 until there is maybe a report one time that the bug is found. I already reported the problem to MS via the dev channel, but didn't get feedback (and now since I returned to 20H2, I likely never will).

 

Last Bios update was this:

Version
F.29
Date
17.07.2020
 
However I cannot tell whether the problem came with that BIOS update or not, I noticed the dwm problem maybe 4-6 weeks ago the first time
TurboTacho
Novice
5,133 Views

Good day to all,

I'd like to add my voice into this issue as well. Since I've been having the same problem as Cody since day 1 of getting my new laptop. I had sourced high and low through various support forums and websites for a definite solution but to no avail nor any noticeable success.

I have attached the relevant SSU data and a screenshot for additional reference.

Capture.PNG

Taking cues from Andrew's questions as a base reference for streamlining the issue:

1- How often does the issue occur? Is it triggered when running any specific app or software?
The issue particularly applies every time when the system is booted up. There is no specific app or software that triggers this, through 2 to 3 hours of continuous use it will start to accumulate memory and bloat up to at least 2-3GB or even up to 10GB.

2- When the issue started, was there any recent change? (for instance: Windows® updates or graphics driver update?)
The above scenario occurred for both Windows 10 2004 and 20H2 (current), I have conducted a clean installation (deletion of driver software -> reboot -> re-installation of drivers). I've used every public release of the driver (i.e. builds 7985/8141/8190... ...8935) and it seems to be displaying the same issue.

3- Have you tested on a different system with Intel® Graphics? If yes, what are the CPU/Graphics model devices and the behavior?
I have another Intel-powered laptop. Running on an i7-7600U and HD 620 on Windows 20H2. The DWM runs well below 100MB and no memory leak.

4- [*Trimmed for relevance*] Do you have additional references regarding this statement? (so we can have more visibility of the behavior and impact of this).
This system is running on both Intel UHD 620 and Nvidia RTX 2080 Max-Q (version 457.09), as it's a gaming system. The RTX graphics has no issue as it's disabled under normal use conditions (i.e. web-browsing, YouTube, emails).

The system originally shipped with build 7642 as stock but I swiftly took the decision to do a clean upgrade on day 1, hence I'm unable to fully determine if the DWM issue persists.

Restarting the iGPU driver via the hotkeys doesn't resolve the issue, unless DWM process is killed and restarted, which briefly resolves this until few hours later...

I've also tested enabling/disabling hardware accelerated GPU scheduling, and it doesn't resolve the issue.

5- Please provide step-by-step instructions to replicate this behavior so we may try to test it.
I do not have any specific step-by-step instructions to replicate this problem, as it occurs randomly at time.

 

I hope these information would be valuable to the relevant teams analyzing and troubleshooting the issue.

Cheers

5,127 Views

Just for curious, how about disable hibernate?
Here is the link for disable it:https://docs.microsoft.com/en-us/troubleshoot/windows-client/deployment/disable-and-re-enable-hibern...
Specific File (Hiberfil.sys) may involved in this issue, and disable hibernate will also disable fast startup.
BlueGuitarCL
Beginner
4,932 Views

I want to make a little input on the bug, cause i've been dealing with it from more than a month restarting dwm via processexplorer64, but i normally shutdown the pc with fast boot deactivated but the laptop takes a lot of time booting up (~10m or more). I deactivated hibernation too so i can check if that helps with the booting speed.

Also, i'll post here some pictures of how many memory dwm allocates on both ram and swap of different days i have saved on my laptop.

My laptop (Acer Aspire A515-51G) details:

  • Windows 10 Home Single Language v2004 10.0.19041(.448?)
  • Processor: i5-8240U @ 1.60GHz
  • Bios and motherboard info on a screenshot
  • 1x8GB ram, 13,6GB page/swapfile
  • Dedicated Graphics card Nvidia GeForce MX130

For this bug i had to install Rainmeter+SysDash for monitoring ram/swap usage always before my pc get stuck. It started to happen after someone told me to use Intel Driver Assistant for recieving driver updates.

Przemyslaw
Novice
4,728 Views

Hi,

I am also experiencing this bug. My machine is Acer Helios 300 (i7-9750H) with Intel UHD 630 graphics and Geforce GTX 1660 secondary card and 32GB of RAM. However, the Geforce stays unused. The system version is 20H2 build 19042.630, definitely NOT a preview one. The bug started to occur quite recently, I cannot distinguish if it was related to updating Windows to 20H2 or related to updating the Intel drivers - their version is 27.20.100.8681. I do not install anything manually, all comes from Windows Update. The Windows installation is vanilla, not the one installed by the OEM. I did not update the BIOS, neither did I anything unusual. Some extra observations that may add something:

- My working set size of dwm.exe is between 2,0 - 2,5 GB
- I do not use hibernation
- The machine is rarely restarted, I use sleep very often
- This week it was restarted on Saturday and the memory leak occured on Tuesday. Before Tuesday it was ca 150MB.
- It might have something to do with sleep. I am monitoring the working set size with Performance Monitor and found out, that the memory increased some time after the wake up, but didn't catch the exact moment. On the other hand, by then it had already completed a few other sleep-wakeup cycles (Sunday, Monday) before, and it was fine.
- I use the screen saver. That's unusual, but not sure if there is a relationship. It is standard Windows Text3D screensaver.
- Some time before I observed the 2GB usage, I launched modern UWP remote desktop client application. Not sure if there is any relationship. Maybe something after launching UWP apps.
- I use classic mstsc RDP client very often, probably this is not relevant.

TurboTacho
Novice
4,700 Views

I've noticed that majority of the posters in this thread have Nvidia GPUs. I was wondering could it be something to do with the GPU switching or some kind of parallel rendering that's leading to the massive leak?

Unless if the Intel driver is not integrating well with the Nvidia Optimus setting, just some wild intuition running in my mind... Based on what I've read online, it seems that the Intel driver is in full charge of the switching depending on GPU demand per application.

Cody
New Contributor I
4,685 Views

@TurboTacho 

This was also my suspicion earlier, but disabling nvidia drivers didn't make it better. However there might be other nvidia software running too that cannot be disabled, interfering with Intel GPU and thus cause this bug.

Maybe intel lab has a computer with a double gpu (intel+nvidia) to test?

TurboTacho
Novice
4,466 Views

@Cody Yes, seems like the non-dedicated GPU users are also having this issue as well.

 

Anyway, I've tried several ways to replicate this DWM issue in a much easier manner. Turns out it's actually possible in certain ways.

Earlier on I tried creating a few video projects via the Microsoft Photos app. The projects consist of a simple picture and video slideshow with music, about 8 to 12 minutes long each. The export process was sufficient to indicate and trigger the memory leak in DWM without delay. (I additionally confirm that Nvidia GPU wasn't activated during the exporting process at any time.)

End result: 17GB memory consumption after 4 simple video projects completed.

TurboTacho_0-1606238059963.png

TurboTacho_1-1606238126825.png

I've also noticed that in most occasions, this memory leak occurs when I use the 'Shut down' button in Windows, and then booting up the laptop. If I were to restart the system, the memory leak issue is briefly resolved for about 4-6 hours upon logging in.

I strongly believe the hybrid shutdown or Fast Startup might be one of the cause of this issue. Definitely it is frustrating when the leak isn't fully addressed, but could only be temporarily resolved through workarounds which may not be feasible in the long run.

smithy
Novice
4,421 Views

So I ran VMMap by sysinternals on DWM.exe.

Started at 45mb, then 70mb, then 135mb, slowly creeping up.

I then did a hibernate/sleep, then powered up again, massive increase to 285mb. It seems that in the memory, the image data stays around 43mb, but the private data and heap are the culprits.

Not so much private data but the heap.

If you look at the screenshot below, it seems it keeps making strings in the heap of size 16,188kb.

The amount of instances of 16,188kb keeps increasing leading to DWM.exe taking up more RAM.

This is beyond my RAM/memory knowledge, but need someone to work out what is causing this increasing in commit, I assume its not flushing the old unused ram out, so keeps ever increasing...

Anyone knows how to do this? Intel surely knows how?

 

Reply