Why is Explorer.exe leaking VRAM?

[Ren Heika]

So I've noticed on my laptop that Explorer.exe leaks VRAM. I am on Windows 10 and using an Nvidia card. However the Nvidia card is a mobile Kepler Quadro so its using driver 426. I have tested running the system with hybrid graphics on and using the Intel graphics fixes the VRAM leak but I would rather not pipe everything through the integrated graphics. My question is, is there any way to fix the issue or is it just an issue with those Nvidia drivers and I'm going to have to use the Intel?

[danalog]

What laptop model? 
On my Precision M6700 (i7-3840QM, Quadro K4000, same driver release) I have turned on Nvidia Optimus in the BIOS and set explorer.exe (in C:\Windows) to use the iGPU in Nvidia control panel.

[Ren Heika]

24 minutes ago, Mel0n. said:

What laptop model? 
On my Precision M6700 (i7-3840QM, Quadro K4000, same driver release) I have turned on Nvidia Optimus in the BIOS and set explorer.exe (in C:\Windows) to use the iGPU in Nvidia control panel.

Its an HP Zbook, and yeah using the iGPU fixes the issue but I would rather keep the iGPU off, my system has a Mux switch which allows me to run the system just off the Nvidia chip which if I do that causes the memory leak.

[Astyanax]

I've been digging into this for some  time, there is an object leak occuring in the desktop explorer.exe process on some machines that is antagonised by things like DPI changes, it is GPU driver agnostic but having the shell on a different gpu would mask the issue by preventing it from affecting game function.

 

I have already contacted Renton about it on reddit, but if anyone else is seeing the same thing can they list things such as their configuration on the desktop, shell extension and software used, whether or not it is an upgrade from 7 or 8, etc.

 

Theres a missing variable here as the issue still replicates on a diagnostic startup whilst not being present at all on a second machine.

 

Someone elses resolution involved changing the default image handler (which also changes the in use shell thumbnail provider), but i've had no luck there thus far. have also gone through and disabled all shell extensions too.

[Astyanax]

should be fixed on nvidia driver 526

[ChalupaKing]

Started noticing this too on my Windows 10 AMD desktop and my RX 5600 XT GPU. If I log off and log back in the VRAM resets for a few hours. Saves me from having to reboot the PC. Not sure it is completely driver related I'll keep watching it.

[Astyanax]

On 1/27/2023 at 5:00 AM, ChalupaKing said:

Started noticing this too on my Windows 10 AMD desktop and my RX 5600 XT GPU. If I log off and log back in the VRAM resets for a few hours. Saves me from having to reboot the PC. Not sure it is completely driver related I'll keep watching it.

Just to confirm and give heads up, this has never been fixed on Windows 10 1904x and 11 21000, it does involve DXGI object leakage relevant to the windows DPI implementation within Explorer.exe, and is antagonised when used with displays that disappear or change their edid state on sleep or power off.

 

I was provided an estimated regression window of when the issue was introduced, of somewhere around the middle of 17763 patch cycle, meaning the issue has been present and unfixed since 1809 at least.

 

The second machine i mentioned was able to trigger the issue only after setting a dpi exceeding 150% (175+ atleast), DXGI objects would leak periodically till the machine was non-resumable from display sleep, and there was a large amount of of the gpu shared memory in use.

 

it is fixed under Windows 11 22h2 and remains fixed on 24h2.

[Verte]

Having this issue currently. Happens once or twice a day. Just have to reset explorer.exe and I'm good for another 5 or 6 hours.

 

- 9800x3d

- RTX 3070

 

One thing I will note. When I upgraded my mobo/cpu/ram a couple weeks ago. I did not do a fresh install of Windows.