rtx 3090-ti black screen crash

[tahunua]

ok so just this week I installed my brand new RTX-3090-TI and it's got some growing pains.

when attempting to record some gameplay from Fallout 4 using OBS, I experienced a strange crash where the screens go black, but the system apparently keeps running and mouse clicks cause the little windows error sound, so there's enough background processes that the OS doesn't completely lose functionality. however, ctrl-alt-del, win-shift-s, alt-f4, and win-tab don't seem to do anything and I'm forced to perform a hard reboot.

windows system logs don't register anything other than the hard reboots and application logs show the following error (but not with every occurence)

Faulting application name: dwm.exe, version: 10.0.22000.1, time stamp: 0x7cbe2305
Faulting module name: dwmcore.dll, version: 10.0.22000.778, time stamp: 0x9ab5d260
Exception code: 0x8898008d
Fault offset: 0x000000000011e68e
Faulting process id: 0x66c
Faulting application start time: 0x01d89eb6df73b9c7
Faulting application path: C:\Windows\system32\dwm.exe
Faulting module path: C:\Windows\system32\dwmcore.dll
Report Id: 6d7bbe0c-65f4-4099-a131-48e645dca441
Faulting package full name:
Faulting package-relative application ID:

the presentation is slightly different each time, with it occurring once while reviewing the collected footage with VLC without F4 or OBS even running, and the most recent loss even carrying over past the hard reboot. I had to pull the power cable and bleed residual power to get it to display again.

this leads me to think it might be an issue with insufficient power. the 3090-ti has the new style power connector and comes with a splitter dongle to accept 3, 8 pin connectors. my current power setup is a Rosewill 1050W gold PSU, using the two cables I was using for my 3080-ti. they are both 12 pin at the PSU, and split into dual 8 pins. if both connectors on a cable are used, do these default to the power output of 6 pins? does anyone know of a good tool to test this, or know for certain whether adding a 3rd cable would fix this issue?

[mimizukari]

3090 Ti are still super new, it's very common for new NVIDIA GPU launches to have some DOA batches until they get stuff sorted out and/or you buy from a brand that stress tested the card extensively.

What you're describing sounds like a PSU issue I had recently though, are you using something like an overclocked 10980XE? that'd overload your power supply with a 3090 Ti afaik.

[tahunua]

I'm sort of an anti-overclocker, I never really felt like the upside outweighed the risks, unless it was a factory overclock option I never felt up to it. i have a stock R9-5950x. the card is a mainline gigabyte, I'm not sure of their stress testing practices but would be surprised if they did.

imediately after posting I had the crash again while attempting to google Rosewill power outputs so I decided to stop being lazy and added a 3rd dedicated PCIE power cable so we will see if that stops or reduces the crashes.

[mimizukari]

Monitor your temperatures before it crashes since it's happening more often now(?), what temps are your CPU, PSU and GPU reaching right before it happens(next time it does[hopefully you have multiple monitors?])?

I will say that when this was happening for me on my previous PC it was definitely a PSU issue, but a different friend who had same issue fixed it by reapplying thermal paste on their CPU and dusting out so their GPU doesn't overheat/etc.

[tahunua]

ok I have once again suffered the same crash with same windows application error associated with it. this time the computer was sitting idle with no applications running.

I have been monitoring temps and nothing is peaking above 50C.

at this point I think i might be chasing ghosts. I have been assuming it is a problem with my new GPU because this just started a few days after changing out my video cards and the system has been relatively stable until this. however, the fact that this is occuring even when the system is at idle means this can't possibly be an issue with insufficient power as there wouldn't have been any peak power consumption by the GPU at idle.

thinking a little more critically, DWM.exe is the windows desktop manager and is directly tied to the windows GUI so it would make sense that windows would remain semi-functional without any gui displayed if this application is crashing. this would appear to be confirmed as windows update says it last checked for updates half an hour after the last crash while it was still in its "black screen mode". I am going to attempt a DISM followed by a chkdsk and end with an SFC to verify it isn't the OS itself. (I know chkdsk has nothing to do with the OS but it's just sort of a habit to run all 3 of those in that specific order)

[tahunua]

UPDATE:
DISM did not detect any problems.
chkdsk found some minor unused elements that were removed.
sfc found some corrupted driver files but after digging into them it appears that all were related to bluetooth which I don't use on a regular basis.

I suffered another crash which resulted in an automatic reboot (which it has never done before) and this one resulted in a number of windows logs, including a system log entry for a bugcheck 0x116 which is usually attributed to a timed out GPU.

I did change the refresh rate for my monitors from 60Hz to 120 a couple days before errors started so I tried setting those back to 60, which I thought solved the issue because I was able to mess around for a couple hours but I suffered yet another crash.

so I'm kindof running out of ideas here.

[W!cked]

Return & get another.... Went through this with one myself, i've always won the lottery on PC parts but this GPU is just a different type of beast!! I changed cords,ram,psu, the whole works...Finally sent back the GPU and haven't had an issue since getting the new one.

[tahunua]

so I went forward with an RMA on the GPU. after sitting on it for a week they updated the notes this morning that their tests were all normal, so it looks like they will be shipping it back to me in as-is condition. I'm at my wits end here.

unless someone has any other suggestions, I think I am going to attempt to return it for a refund. it has not proven worth the hassle, and my 3080-Ti has been solid in the interim.

[mimizukari]

3080 ti/3090 are very solid, i never trust the lottery with new GPUs, rather wait for all the binned and tested ones to come out a year later give or take... GPUs are the thing that i have the most trouble with if they're a year 1 release.

Although if they tested and it's fine, then it's something with your other hardware?

[tahunua]

Tried both of the solutions in that document. Neither had any effect

[AmaiAmai]

1. Make sure your power supply can actually run your GPU. Calculate your power consumption based on EVERY part you are using in your PC.

2. Make sure your drivers are fully uninstalled from the previous graphics card, and the current one...then reinstall it. You can use Display Driver Uninstaller (DDU) utility to completely delete your driver

3. Before reinstalling driver, make sure you set Windows to NOT install driver updates from Windows Update, and for Windows NOT to update your hardware's firmware from Windows Update because drivers from there are either old or buggy drivers. By default PCs will load up those drivers from Windows Update even if they are older ...

4. Don't think it needs to be said, but you have to be using Windows 10 or later for that GPU, if not, there's your problem. Windows 7 and 8, 8.1 do not officially support recent hardware, even if it runs on them. If you are using one of those operating systems it could be a deeper incompatibility issue, and that is one you will have to either patch yourself or move on to Windows 10+.

5. Try setting your GPU as the primary graphics adapter. You can do this by using Nvidia Control Panel and setting the default GPU to your Nvidia card. You can also disable the iGPU on your CPU from the BIOS on most dekstop or server motherboards (but not all laptops). Switchable graphics sometimes causes instability.

6. If it still crashes, try disabling hardware accelerated GPU enhancements in Windows, see here for how to turn it on, opposite way turns it off: https://www.makeuseof.com/windows-10-gpu-hardware-scheduling-worth-turning-on/

7. If it's a dual-bios card OR your motherboard supports dual BIOS, make sure your Windows install is in UEFI because older BIOS may not be compatible with certain features of the card (R.Bar, compute, DMA, virtualization, etc.).

If that doesn't work, maybe it's best to ask Nvidia or card manufacturer.

[tahunua]

Unfortunately not. I don't have mso installed so it wouldn't have been overwritten. This is a brand new build on a new ssd and chkdsk showed no bad sectors, sfc and dism also rule out corruption of the dwm.dll. also the system is stable with a 3080 ti installed so the issue isn't related to missing or corrupted files.