dwm.exe (Desktop-Window Manager) uses high memory with Intel Intel HD Graphics 630. This problem has been reported before:
Here's a screenshot of my dwm ram usage after 2 hours (1.6 GB):
Please, this bug is already known for 3-4 months, and I do not want to reinstall a 2019 driver...
No progress has been made since we haven't been able to find a way to reproduce this issue in our lab using different platforms (laptops, NUCs with and without Hybrid graphics) and all the input you have provided so far.
I would recommend reporting this issue to Microsoft. Usually if they find a fault in the OS component (dwm.exe) linked with the Intel driver then they will work directly with Intel to address it.
I have this issue too on my NUC8i7bek (Iris Plus 655 graphics) with fully up-to-date Windows 10 and currently the latest beta driver 18.104.22.16877, though every driver version I have tried has been affected. My system also has a 4K monitor (Dell U2720q, specifically). The hardware is important here, because I cannot reproduce the issue on a NUC6i5syk (Iris 540) with a 1080p TV and the same driver and OS version.
I do have a method of reproducing this issue 100% of the time. My system, for the record, has fast startup and hibernation both disabled, and I never use sleep. It is configured to turn the monitor off (and lock the screen) after 5 minutes of inactivity. If I press Win+L to lock the screen manually, the monitor shuts off within a minute or so.
To reproduce the issue, I reboot my system and log in. Then I lock the screen and wait for it to shut off the monitor automatically. Once it does, I press a key to turn the screen back on, log in again, and repeat the process. The second time the monitor comes back on, if I perform almost any action, DWM starts leaking memory very rapidly. This includes opening/closing windows, moving windows, scrolling within windows (browsers, both Firefox and Chrome, seem to make it leak the fastest), and watching videos within a browser all cause leakage. Then, if I lock the screen again, wait for it to shut off, and wake it up a third time, it stops leaking RAM (though what was previously leaked is not freed). The leakage will occur again after some number of additional monitor sleep cycles, but whenever it is leaking, a sleep cycle will temporarily stop the leak.
I have also found that if I use SysInternals RAMMap to "Empty Working Sets" followed by "Empty Modified Page List", the leaked RAM is freed. This does not stop the leaking, however, if it was actively occurring.
I've had a memory leak in the Desktop Windows Manager and have been unable to allieviate it. I tried updating to the next version of windows as it was offered through Windows update but not yet mandatory, and to no avail. I've run DSIM and SFC but even when they do repair files it doesn't seem to change my memory situatuon. Saw a thread on the Microsoft forums where somebody reccomended updating Intel HD graphics drivers but nope, nothing. Tried disabling all background apps and programs on startup but still nothing. On a blank 1080p desktop with nothing but task manager open I can sometimes see up to 10 of my 12GB of memory used by DMW.exe and I've done everything short of making a repair disk or reinstalling Windows that I can think of. The 9nly thing that will bring my memory usage down is a restart but I can't just reset a few times every day, especially since the rate at which the memory leak happens seems unpredictable.
SOLUTION: The Intel HD drivers are at fault. As I was reccomended in the comments below, download the intel HD 22.214.171.12486 drivers,download Wagnard's DDU, install it, boot in to safe mode and use it to remove the intel drivers, reboot as recommended and install the intel drivers, restart, shut down, then boot in to Windows normally a your issues should be solve.
Bonjour à toutes et à tous
Tous d’abord bonne et heureuse année 2021 ainsi qu’une bonne santé
Excusé de mon langage je suis Français.
Je suis tombé sur cette discussion suite au même problème avec un notebook
le cpu est un celeron j4115 avec intel UHD 600 en support vidéo
j’ai le notebook depuis une semaine quand je l’ai reçu il était sous Windows 10 pro 1903
Tous allez bien, passez en 1909 avec Windows Update
Tous allez bien, passez en 20H2 avec Windows Update
Depuis même problème avec dmw.exe
En suivant les guides j’ai donc mis la version 126.96.36.19985 et tous va bien
j’ai désactivé la mise à jour automatique dans :
Paramètres avancés du système > Matériel > Paramètres d’installation des périphériques
Donc il y à bien un problème avec les derniers drivers Intel et Windows 20H2
PS : Mon notebook est un Kuu K2s sous Windows 10 pro
I experienced the dwm.exe memory leak problem with versions 188.8.131.5253 up to 184.108.40.20626 inclusive.
However, I tried the older 220.127.116.1183 version and so far have not had a problem with the dwm.exe memory leak.
Hello! I have the same bug on my Lenovo Carbon Gen8 (intel UHD 620). After hybernation my dwm.exe ate 9,8 GB. OS is Windows 10 20H2. I updated intel drivers (to 18.104.22.16826) and will continue to monitor the issue to collect a report. I didn't have any issues on previous driver version until I activated hybernation (though I might just haven't noticed).
I'm quite disappointed with intel team response "we couldn't reproduce it so it's not our problem". Hope it will be resolved soon.
Since the issue only occurs on driver versions 27.XX.XXXX and higher, I'd assume it's something to do with the WDDM 2.7, which makes sense as it was implemented early 2020.
I'm not sure if it helps, but on 22.214.171.12426 (and every driver > 126.96.36.19953) I encountered strange bugs in some games. The most reproduceable is Rimworld color coded bars (in the schedule menu, pawn's needs menu), all bars are gray, while on all previous drivers including 188.8.131.5253 colors are displayed correctly (blue, red, yellow, green etc.). Official forum link: https://ludeon.com/forums/index.php?topic=53555.0. I downgraded drivers to 184.108.40.20653 to fix this issue, but now dwm.exe leaks memory again (up to 700 Mb in 1h after os start with fast boot, ~5-10 Mb/h after dwm.exe restart).
It seems I don't have any choice but to rollback to 7985 driver until Intel/Microsoft figure this out.
I don't know if it helps, but I encountered strange bugs in some games on 220.127.116.1126 with Intel UHD 620 (and on any driver > 18.104.22.16853), though dwm.exe leak was minimal. For example. in Rimworld color coded bars are all displayed in gray color (in schedule menu, pawn needs menu). I downgraded to 22.214.171.12483 and the problem was fixed. It seems some links aren't allowed here, but you can find the topic on official Rimworld forum by searching "ludeon 53555.0" on google. On 126.96.36.19953 dwm.exe started to leak again (up to 700 Mb in 1h after start with fast boot, 5-10 Mb/h after dwm.exe reset).
It seems I don't have any other choice but to rollback to 188.8.131.5285 until Microsoft/Intel figure this out.
I was able to reproduce the dwm.exe memory leak using version 184.108.40.20626 on an Intel UHD 630.
Once the memory usage started to go above about 300Mb, it didn't come back down, even after all Apps were closed. I tried this workaround, and after a couple of days of regular use I haven't see it use much more than 100Mb. The memory usage also drops regularly, and returns to "normal" values when all Apps are closed. I'm using Windows 20H2 Build 19042.746 on a Desktop PC, so it may not work on a Laptop or other versions of Windows 10?
Go to "Graphics performance preference" here -
Settings -> Display -> Graphic Settings -> click on the Browse" button and select dwe.exe in the System32 folder. Click on the "Options" button and select "Let Windows Decide" (I think this option is marked "System Default" on some versions of Windows?).
I tried both the "Power saving" and "High performance" options, and they did nothing, other than possibly making it worse!
I've been using my PC with this configuration for about 7 hours today, and most of yesterday, and the current dwm.exe memory usage is 100Mb. Of course I can't guarantee that the memory problem won't come back again? I can't guarantee that it'll work for you either?
I found in a thread here:
TurboTacho: We are aware that the Intel graphics driver 220.127.116.1185 is the last known best working set that had no such memory leak issue.
I am also kind of programming stuff, so I do know that bugs, that aren't reproducable are nearly impossible to fix, also I heard that this is a heap problem which belong to the hardest nuts...
However if it is known that 18.104.22.16885 is the last version without this bug, many seem to have (I hope this version is right, since I didn't test it myself), @intel can I recommend to maybe look into the protocolled changes to the next version to maybe figure out which change could have caused this bug?
I would also happily test any experimental fixes
please discard/ignore above message. I systematically tested all previous drivers available at
myself and can tell you exactly the last stable version without the dwm leak:
It is 22.214.171.12476
as soon I update to the next version (126.96.36.19987) the dwm leak happens on my computer... maybe you can check the change log between both versions and hopefully find something obvious that may cause this bug 🙂
I checked the Microsoft update catalog for intermediate versions.
There are v8528 and v8531 for DG1 only and come with a new kernel mode layer handler.
My best guess is that an exclusive tweak for the new kernel driver stranded in the old one 🤔
Hmm. I think it depends on Intel, I happily test more versions if this helps. Like if they can provide 8520 and this bugs I continue with 8495, if bugfree I can test 8545 or so. This way we could easily find the culprit update.
I had the same issue here with windows 2004, CPU I7-9700K with UHD 630 and no dedicated card. dwm.exe used to ate up 20GB of my 64GB memory after hoours to days of using, which was serious while I am working in memory-intense environment. Can confirm the problem disappeared after downgrading to 10. 188.8.131.5285. Hope Intel won't remove this version of drive from the website before this issue is fixed.
After one week with Windows 10 build 21292 and still the driver v184.108.40.20616 I don't saw this problem. It could be luck, but I have a little bit hope that it is fixed by Microsoft. Today I expect the next Win10 Dev build and I will keep an eye on it.