Strange Desktop glitches w/3090

L

Lamont

Supreme [H]ardness
Joined
Oct 26, 2004
Messages
4,252
I lately (before the driver and after the recent update) I have been getting this strange glitch on the desktop background. It only affects the background so I won't know until I close an application or window. The glitch is not consistent, because I can go a day or so and not see it, but when it happens, it just grows and grows in roughly the same horizontal area on the screen. Attached is a small video of the issue along with I hope is relevant information.

Maybe connected to this issue, but I can't play many games and I get crashes and horrible performance.

2021-10-07 10_12_27-Window.png2021-10-07 10_18_26-Window.png2021-10-07 10_18_32-Window.png2021-10-07 10_18_37-Window.png2021-10-07 10_18_47-Window.png2021-10-07 10_18_58-Window.png
 
You can troubleshoot this by:
  • Underclock the memory using MSI Afterburner -> If the artefacts go away, you have GPU memory issues
  • Underclock the GPU
  • Put card in another computer and test
    • If good then try DDU and reload drivers, try different drivers
    • If problem follows to the other computer -> RMA
My guess would be vram corruption but can change the color settings to see if that makes a difference:
  • Open up Nvidia Control Panel
  • Go to Display -> Change resolution
  • Use NVIDIA color settings -> Try out different combo's
 
Thanks, I did some tests when I got home, it's a card issue. It starts to glitch out after running to 50c. RMA time. Hopefully they can do an exchange, if not I can run my 980ti again while waiting.

Thank you both again.
 
Updating this as I did more tests that make me feel it is not a GPU issue, but acts like a GPU issue.
  • So I did tests (memtest, Win10Mem test, underclocked as noko mentioned, tried another system) and nothing came up, so I held on to the card as it was acting normal.
  • Just recently I played Wolfenstine: The New Colossus, got frequent crashes.
  • Swapped out the 3090 for my 980ti and when I played Wolfenstine Youngblood, I got the same types of crashes with the 3090.
  • Ran memtest, windows own memory test and nothing came up.
However something did happen twice: I came back from one of the Furmark tests and the computer had reset. Same thing happened with OCCT when I did a power test. Ran the tests again and nothing happened. So now I have the feeling it's a different hardware issue. If it were PSU, the machine would have just rest all the time instead of crashing an app...

1643171875452.png
 
That doesn't mean it's not a PSU issue. Typically, really weird issues like this are almost always PSU or the motherboard once you've eliminated the video card itself.

I'd try swapping the PSU first, it's the easiest next step you can take.
 
Been digging into this all weekend with the 980ti...
  • Swapped to a 750w psu.
  • Traced one crash to PhysX.
  • No crashes on Wolfenstine.
Moved to swap the GPU to the 3090 (DDU drivers). While swapping the cards, the 750w was missing a 8 pin (had an 8pin and a 6 pin only), so I looked into the box and could not find it. I moved the Corsair 850 off the desk and I noticed I was using only one PCIe power cable, it had used one that split a single into a 6+2pin and a 8pin. At the time I thought this to be correct.

So now I am prepping to kick myself a few times as I put things back together and start testing again...
 
I spent the week cleaning my workspace to look for the other 8pin for PCIe for the 750w. Found it and restarted the tests.
  • 850w PS crashed on both the 980ti and the 3090.
  • Peak power is about 350w~351w. The app would crash, logs only peaked at 352w. Once I got a BSOD, but this is rare.
  • Moved to the 750w for 980ti and 3090.
  • Peak power was at 360w give or take a w or two at times. Logs showed a max of 364w. No crashes.
Ran a bunch of tests, played a ton of games (ones that would crash like Metro) and no issues at all on the 750w.

Called up Corsair and they set up the RMA process. Smooth process so far. Thanks everyone for the input to figure this out. This is my first buggy PSU in the almost 18 years of making PC's.

So glad it's not the mobo, CPU or GPU.
 
This is the most messed up thing ever...

On a backup PSU. Been gaming, running UE5 and 3DS Max since Sunday. Games are at max settings and keeping the room nice and warm. Installed Sifu Thursday night and started playing...

Crashed to desktop. No clue why it crashed all of a sudden. Started going crazy.

theorymeme.jpg

Ran memtest86 again, started getting errors. Like 400+ errors.
  • Pulled all the ram, did memtest on individual sticks. No errors.
  • Then in pairs. No errors
  • Then different channels in pairs to see if I had a bad slot. No errors.
  • Put in my old kit (4x 8GB), no errors.
Cleared CMOS each time I ran the test.

Then put all the memory back in and ran the tests, got errors. So I set clocks to 3100 instead of Auto. No errors. Memory is DDR4 3200 (Patriot 64GB kit x 2). Gonna game and see how this goes.
 
This sounds very much like bent pins on the CPU socket of your motherboard. Way back when I had similar weird issues with memtest on a motherboard of a customer, it turned out to be the case. Sounds a bit too similar to me.
 
STOP WHAT YOU ARE DOING, AND LISTEN:



just for deductive approach
DDU and install STANDARD DRIVER 472.84


report if it goes away.

WHY?


I’ve reported it to nvidia, posted it on guru3d, etc, there are multiple folks reporting it and the same videos on guru3d

its the dch driver and the 3090
the cause: the new MS and NVIDIA agreement has continual spamming of the DWM service (1000 of times a second) install perf mon and see for yourself, that generates the issues and errors.

just try it -->

l posted a similar video and even had an ram swap, it was the driver series from dch 496 onwards

Its not your video card as the window gets rid of the polygons, and your mouse can go over the polygons without disappearing, lastly its a driver issue and not hardware as
you can do a pc screenshot of the issue and the polygons are displayed in the picture (think about it).

The issue arose from the shift to DCH away from standard driver with 496 dch onwards. There are 11 tickets in for the issues, it is an nvidia / MS issue with the drop from standard.


prove it to yourself: install the bare minimal standard 472 series driver, reboot, watch the issue magically go away.


You don't have to take my word for it, but I do have 4 3090 rtx FE cards that I can reproduce your issue with ease by installing 496 dch, 497 dch, 509 dch, and 511 dch.
If you only install the video driver, and physics in 511.79 it may go away, it goes away for sure if you install STANDARD (non DCH) 466, 471, and 472 drivers


BUT DON'T just take my word for it

Here, these shits were taken and reproducible when 496dch and above, driver is installed.
reverting to standard 466, 471, or 472 resolves the issue.

squares.jpg


this is a screen shot of 2D desktop, now think about it, a screenshot, that means it’s not the video card or the monitor, it means windows is aware of the polygons.
polygons.jpg




2x 3090 RTX Founders Edition & SLI / NvLink bridge newest bios
2x 8280L, 56/112 cores, Asus c621 Sage Dual socket motherboard, bios 6605 modded
1.5 TB ram. DDR4 ECC LRDIMMs
1600W silent digital power supply
(Data drive) 4x VROC Raid 0 Micron 9300 Max (12.8TB each / 51.2 TB volume) and 8TB backup drive array
(OS Drive) Sabrent Rocket 4 Plus (4TB). 4x Phanteks T30,120mm, Noiseblocker 92mm for nvme drivers
Asus PA32UCG-K monitor, 103 firmware
TT SFF case, micro. and heavily modded
MS Data Center 2022 (customized) & Ubuntu (customized)


https://gravitymark.tellusim.com/leaderboard/



587436_IMG_9739.jpg

587455_IMG_8867.jpg


Install 472.84 STANDARD and issue will go away.

Install video driver, physics, and nothing else. (GFE installs the constant ping to the DMW as well)




yes, the devil is in the telemetry details.
 
Last edited:
Venturi said:
STOP WHAT YOU ARE DOING, AND LISTEN:



just for deductive approach
DDU and install STANDARD DRIVER 472.84


report if it goes away.

WHY?


I’ve reported it to nvidia, posted it on guru3d, etc, there are multiple folks reporting it and the same videos on guru3d

its the dch driver and the 3090
the cause: the new MS and NVIDIA agreement has continual spamming of the DWM service (1000 of times a second) install perf mon and see for yourself, that generates the issues and errors.

just try it -->

l posted a similar video and even had an ram swap, it was the driver series from dch 496 onwards

Its not your video card as the window gets rid of the polygons, and your mouse can go over the polygons without disappearing, lastly its a driver issue and not hardware as
you can do a pc screenshot of the issue and the polygons are displayed in the picture (think about it).

The issue arose from the shift to DCH away from standard driver with 496 dch onwards. There are 11 tickets in for the issues, it is an nvidia / MS issue with the drop from standard.


prove it to yourself: install the bare minimal standard 472 series driver, reboot, watch the issue magically go away.


You don't have to take my word for it, but I do have 4 3090 rtx FE cards that I can reproduce your issue with ease by installing 496 dch, 497 dch, 509 dch, and 511 dch.
If you only install the video driver, and physics in 511.79 it may go away, it goes away for sure if you install STANDARD (non DCH) 466, 471, and 472 drivers


BUT DON'T just take my word for it

View attachment 448261
View attachment 448262



2x 3090 RTX Founders Edition & SLI / NvLink bridge newest bios
2x 8280L, 56/112 cores, Asus c621 Sage Dual socket motherboard, bios 6605 modded
1.5 TB ram. DDR4 ECC LRDIMMs
1600W silent digital power supply
(Data drive) 4x VROC Raid 0 Micron 9300 Max (12.8TB each / 51.2 TB volume) and 8TB backup drive array
(OS Drive) Sabrent Rocket 4 Plus (4TB). 4x Phanteks T30,120mm, Noiseblocker 92mm for nvme drivers
Asus PA32UCG-K monitor, 103 firmware
TT SFF case, micro. and heavily modded
MS Data Center 2022 (customized) & Ubuntu (customized)


https://gravitymark.tellusim.com/leaderboard/



View attachment 448264
View attachment 448266

Install 472.84 STANDARD and issue will go away.

Install video driver, physics, and nothing else. (GFE installs the constant ping to the DMW as well)




yes, the devil is in the telemetry details.
Click to expand...
Have not seen this on any of those drivers for an EVGA 3090. Well at least yet. Is this more than just a driver problem? A bios/driver/windows issue or something else also?
 
noko said:
Have not seen this on any of those drivers for an EVGA 3090. Well at least yet. Is this more than just a driver problem? A bios/driver/windows issue or something else also?
Click to expand...

This problem has been reported by others on the 3DCenter siter as well as Guru3D

cause is the one I showed, reproducible

there is a house of cards for how it happens:

driver 496 and above,
DP 1.4 DSC, HDMI 2.1
10 bit colour
fresynch, freesynch pro and some G synch
3840x2160 or any combination
windows 10 or 11

several of these items, combined with a 496 or above driver no a 3090 yields the 2D desktop polygons in windows

the mouse can go over the polygons not not be affected, dragging a window over the s\dots/polygones keeps the dragged window intact like a veil going over the polygons, screenshot can capture the polygons - that way you know it’s not your card or monitor

3dCenter, Guru3D, anandrech forums, have reported similar issues.

cause: dwm spamming from desktop telemetry for ms store appx.



all issues go away if Standard (non dch) driver is installed such as standard 472.84

side note: nvidia replaced/rma for (new) two out of 4 3090 rtx fe before it was discovered that it was the 496 dch (and above) driver


according to the tickets, from dealing with tech support, 11 other tickets resulted in rma cards before the culprit was found. But.…NO DRIVER SOLUTION HAS BEEN ISSUED, other than revert to standard drivers (466, 471, 472). Apparently fixing the issue breaks whatever other agreement they have with MS.

also, issue does not exist in nvidia/Ubuntu installs
 
Last edited:
Venturi said:
This problem has been reported by others on the 3DCenter siter as well as Guru3D

cause is the one I showed, reproducible

there is a house of cards for how it happens:

driver 496 and above,
DP 1.4 DSC, HDMI 2.1
10 bit colour
fresynch, freesynch pro and some G synch
3840x2160 or any combination
windows 10 or 11

several of these items, combined with a 496 or above driver no a 3090 yields the 2D desktop polygons in windows

the mouse can go over the polygons not not be affected, dragging a window over the s\dots/polygones keeps the dragged window intact like a veil going over the polygons, screenshot can capture the polygons - that way you know it’s not your card or monitor

3dCenter, Guru3D, anandrech forums, have reported similar issues.

cause: dwm spamming from desktop telemetry for ms store appx.



all issues go away if Standard (non dch) driver is installed such as standard 472.84

side note: nvidia replaced/rma for (new) two out of 4 3090 rtx fe before it was discovered that it was the 496 dch (and above) driver


according to the tickets, from dealing with tech support, 11 other tickets resulted in rma cards before the culprit was found. But.…NO DRIVER SOLUTION HAS BEEN ISSUED, other than revert to standard drivers (466, 471, 472). Apparently fixing the issue breaks whatever other agreement they have with MS.

also, issue does not exist in nvidia/Ubuntu installs
Click to expand...
I cannot reproduce. Using most current driver, 511.79, EVGA 3090 XC3. 3840x2160 (DSR), 144hz, 10bit, Freesync monitor with GSync forced, with or without HDR. I did loose Nvidia Settings Control panel, after going to 3840x2160 it will no longer open, even going back to native 2560x1440p resolution. Since I cannot test native 4K with the 3090, I can with the 3080Ti, if I get a chance I will do that (wife is using the 4K HDTV, engrossed in old General Hospital episodes nonetheless at VCR resolutions :D.)

Anyways this does not look to be very wide spread, as in thousands or tens of thousands reporting but appears to be an issue with some setups.
 
Venturi - The glitch issues went away with a driver update (using DDU).

So I had pretty much given up as I was under pressure to deliver some assets. After it was finished I went back to looking into the issue.

I got the new PSU and set up my other machine with the 3090 and it ran fine. While the new comp still crashed and was unstable with the 980ti.

Pulled everything apart, got my loupe and took a look for bent pins and damage to the mobo. That was a good 3 hours. Next day my new mobo arrived (MSI Z590).

Put everything in the new mobo and still crashed. Pulled out all the ram and started testing each stick. Each stick passed. Then played the games that usually crash, and they all worked just fine. Then I took a look at the memory compatibility chart. My ram was not listed, and past mobo's had this chart and my ram wasn't listed the majority of the time and never was an issue. What I did notice was that the support for 32GB sticks was spotty and very few, also I noticed that the clock speeds for these kits were 2400. I put all 4 sticks back in, reset the mobo settings with the jumper, set the clock speeds for the ram to 2400. Got back into Windows, ran all the tests (Use Unreal Engine, play Horizon Zero Dawn, Metro Exodus.. etc.) and *bam*, no issues at all. Fans on the GPU don't go balls out in Unreal 5 anymore on simple scenes (or even the Matrix Demo data...).

I can call this a DFO error and a $180 lesson (I am keeping the MSI because I am NOT pulling everything out again...).

Thanks everyone for chimed in with tips on what to look for. Now I can get back to work without having to save my Unreal scenes every minute... now it's every 5 minutes like normal.
 
You must log in or register to reply here.
Back
Top