UE5 gpu crashed or d3d device removed

Ok, to update this: I tried installing MSI Afterburner and undervolting, but that did not resolve the issue either.
I also updated to the latest gameready driver again (531.29), didn´t help either.

Followed some other random advice and:

-Tried changing the PCIE settings in the bios to use gen 3 instead of gen 4
-Tried putting the card into a different slot.

Neither of them help.
I also used furmark again to stresstest the gpu with no errors.

To me it is hard to see how any of this would be hardware related, when it ONLY happens with Unreal engine 5 and direct x12 enabled (from some games forums it seems that directx 12 might also cause similar issues there).
Trying it on two separate workstations with separate PSUs and mainboards and rtx cards and having the same issue also points to that.

It does NOT happen on all scenes, but I have not been able to pin point what makes them different.

And I still find it suspicious that fps drops by exactly half, if you open any editor windows.

1 Like

Ok.
I think I might have pinned down the problem somewhat.
The level that kept instacrashing with GPU removed constantly, had some niagara systems in it.
I don´t think its the niagara system though, I think its the material that is causing it, which is a particles shader with alpha and in add mode…Its not crashing in unlit mode.
Its also not crashing if I hide the static mesh in the scene, which is a hipoly imported mesh.
Doesn´t crash with simple basic shapes, but also crashes with other hipoly mesh, but only in combination with the niagara system.
AFAIR I also had massive crash problems on another scene, where I used a couple of fog cards.

If anyone has an idea what exactly the problem might be and if there is a solution, I´d be happy to hear!

I´m going to try to get a colleague to test the same scene and see if it can be 100% pinned down to that issue.

I believe its also something similar
i have a UE5 project with dx12. dx12 is a requirement of the niagara fluid system im using. if i use dx11 then no crashes. it has something to do with niagara

I don’t use Niagara, even on empty project - it crashes.

Just install the Nvidia 531.41 drivers (Game Ready or Studio) and get “GPU Crashed or D3D Device Removed” every 2 minutes! This issue with Windows 11 and DX12 has been around for months, and nothing has been fixed since. The various tips found on the forums work more or less well. For me, the 517.40 drivers are much more stable, so there is a problem with the drivers.

It’s time for Epic / Nvidia / Microsoft to fix the problem once and for all. With the latest Nvidia drivers, Unreal becomes completely unusable for me!

2 Likes

I don´t think its niagara, but it might have to do with translucency/masked materials.

1 Like

Are there any caveats in rolling back to 517.40?
I don´t quite remember why I didn´t reconsider this option, as I remember trying it before.
Is it maybe because it doesn´t support SM6 or something like that and Unreal 5.1 requires that?

I´m doing some more digging right now, one thing thats not been mentioned before is Monitor setup:

For those with crashing:
Do you all have multiple monitor setups? I think I also tried physically disconnecting them, but that didn´t do anything, but I´m not sure.
Also, one of my monitors is can only do 60hz and another one can only do 59.97, so I´m running two on 60hz and one on 59.97hz.

I did some more testing with the niagara system though and turns out I can pretty reliably crash the GPU on a test scene as well, just by dragging it into the level makes it crash after a few seconds.

Gonna do some more investigating to see if I can pin it down further.
Because I´m pretty sure I also got these crashes in a scene with no Niagara in it.

single monitor setup for me. unless a vr headset counts as a separate monitor

Let us know if this fixes the issue

Give it a try, I have no problems with the 517.40 drivers and the support of SM6 or others. the latest drivers on my side are crashing unreal every minute, so there is definitely something playing in the drivers. I read on the forums that the gpu crash problem really started from the 52x.xx drivers.

Greetings. Using 531.41 Studio,
Using millions of particles advected by realtime fluids runs slow but does not crash.
Zooming camera into Metahuman in viewport crashes it with 100% success rate. Will try the 517 driver solution.

I suspect the crash occurs on when I zoom past some threshold that immediately tries to load ALL THE TEXTURES at once at fails. Will try to create a new project with VTs and VT import by default. (read more for test results)

Enabling VTs and doing that “zoom” thing gives me red errors about getting over the budget in the top left corner of viewport. My hunch was right. Enabling RT for Lumen and shadows crashes it with 50/50 chance, still.

========
So anyway, the results of testing:

531.41 Studio: Crashing the “zoom into metahuman” on Epic preset 100%, on High scalability not immediately, but eventually.

517 studio: Less crashing the “zoom into metahuman” both Epic and High. High seems to be the most stable of them all. VTs on, import textures as VTs on, use RayTracing OFF. Even having this on while other Lumen/Shadow things disabled makes it (much more) unstable!

531.41 Studio: Rendering another scene: thousands of animated megascan foliage, big environment with 8k textures, RT on, RT Shadows ON, all lumen RT on top settings - working perfectly.

The best guess culprit for Metahuman crash: You zoom in close enough it triggers something wonky - tried deleting the peach fuzz, it does not help (i first concluded it was doing some sort of a glitchy switch upclose, but no luck). Flying through head and back again pretty much gives 100% crashrate.

Enabling VTs, and VT import by default helps. Scalability to “High” helps. Disabling HWRT as a fact in rendering settings helps. MetaHumans seem to have problematic setup on import, I would have to ask dear UE devs to take a look at my findings.

The results are Scene dependent!

Like i said, interactive millions of particles in Niagara works fine (just slow, but expected). Having a heavy environment scene with all possible HWRT things enabled works fine (low fps but it chugs along just fine). This is all on the allegedly crash-happy Studio 531.41 drivers. Those drivers work just fine! (reinstalled 3 times back and again just to be sure, and yes i rebooted after each reinstall, including the 531 and 517 ones)
Possibly the most bizarre finding is that if you do not zoom into MH very close, you can enable all RT, bells and whistles, RT shadows and everything, and it will work perfectly. Right up until you want to get a bit closer.

I’d love developers to take a look at this problem, it is persistent, in fact so consistent it’s almost a feature. Jokes aside, i hope it gets fixed sooner than later. Thanks everyone. myself I’m done doing testing and reinstalling drivers at least for a week haha.

1 Like

Interesting: this problem does not happen in the Sequencer, even if i force-fly the Cine Camera actor through the metahuman.
:face_with_peeking_eye: curiouser and curiouser

This has become essentially a blocker to continue development for us.

I’ve tried the latest studio drivers, as well as the 517.40 drivers mentioned above. Same result. We’re on 5.1.1, things were more stable before 5.1, but less stable since 4.27.2. I hadn’t experienced these crashes prior to 4.27.2.

I’ve tried every remedy proposed on the Discord server and forums (regedit, downgrade drivers, upgrade drivers, fresh Unreal install). The only thing I haven’t done yet is move back to Windows 10, but this would be such a pain to try that I honestly don’t see it as a potential solution.

I don’t understand how this has plagued multiple people for months and nobody has done anything about it? It seems Epic say it’s on Nvidia, and vice versa. I’m running a 3080ti here that is only about 6-8 months old and I’m struggling to keep the engine open for more than 30 minutes at a time while doing anything with more than a few hundred Nanite meshes/materials in view. Blueprint work in an almost empty level is about all I can do at the moment.

I realize this is just an unhelpful rant, but I’m so frustrated by this, and not knowing what else to do next.

6 Likes

I can totally feel you on all of this!

Has anyone created an official Unreal issue about this?

1 Like

i keep pressing “Send and restart” or “Send and close” every time it crashes (very often), so pretty sure they know

Give it a try, I have no problems with the 517.40 drivers and the support of SM6 or others. the latest drivers on my side are crashing unreal every minute, so there is definitely something playing in the drivers. I read on the forums that the gpu crash problem really started from the 52x.xx drivers.

Well, as the next poster stated: Doesn´t seem to fix it, only makes it less prominent.
I might still try it out.

Its not metahumans either, since it also crashes on other scenes.

And its not windows 11 related, plenty of Windows 10 users here with the same crash (including me)

Limiting framerate is still the most consistent “fix” for it, with the caveat that you are severely limiting your editor FPS, if you work on demanding scenes (because apparently the performance scales on the limited FPS).

Any one have a solution? I usually have crash at alt tab.

I have this problem on Windows 10 too - literally just selecting objects in the viewport or opening quixel bridge or even just opening texture window crashes. 3080

Tried downgrading to Windows 10, and no change for me.