Multi Threading DXGI Device Removed Error

[Paraso]

I've looked through countless threads regarding this issue and all of the solutions I've tried so far either don't apply to me or haven't worked. I'm experiencing a DXGI device removed crash on OB MT within 20 minutes of spawning into a mission. I have never experienced this error/crash in non-MT OB nor any consistent GPU-based error/crash in any other game I've played.

Log from today (6/24) was using the ryzen high performance power plan, featuring only one DXGI error near the bottom.
Log from (6/09) was using the ryzen balanced performance plan, with many DXGI errors.

Neither setting "TdrLevel" to 0 nor setting any delay in registry editor have ever given positive results, however for the 6/24 log neither of those settings were actually in place. I don't have any overclocking or volting changes, I've yet to try any of the default windows power plans, setting a custom pagefile also has not yielded any positive results though again that was not set in the case of the 6/24 log. Custom RAM profile is enabled in the BIOS though my RAM is on the slower side (2400mhz). I don't have Volanta's plugin for DCS nor aorus engine.

dcs.log-20230609-034157.zip dcs.log-20230624-215259.zip

[Flappie]

Hi. You're having the "DX device removed" symptom. It means DCS cannot reach GPU anymore at some point.

2023-06-09 03:37:51.945 INFO    EDTERRAINGRAPHICS41 (13092): surface5 gc() 5.327500 ms
2023-06-09 03:41:29.654 WARNING DX11BACKEND (13092): Can't map structured buffer.
2023-06-09 03:41:29.654 ERROR   DX11BACKEND (13092): Failed assert `0` at Projects\render\dx11backend_win8\Source\DX11StructuredBuffer.cpp:100
2023-06-09 03:41:29.656 ERROR   DX11BACKEND (13092): DX device removed. Reason: 0x887A0006
2023-06-09 03:41:29.657 WARNING DX11BACKEND (13092): Can't map structured buffer.
2023-06-09 03:41:29.657 ERROR   DX11BACKEND (13092): Failed assert `0` at Projects\render\dx11backend_win8\Source\DX11StructuredBuffer.cpp:100
2023-06-09 03:41:29.658 ERROR   DX11BACKEND (13092): DX device removed. Reason: 0x887A0006
2023-06-09 03:41:29.662 WARNING DX11BACKEND (13092): Can't map structured buffer.
2023-06-09 03:41:29.662 ERROR   DX11BACKEND (13092): Failed assert `0` at Projects\render\dx11backend_w

This can be caused by many things. Try these solutions:

• Ensure Windows power plan setting is correct (default or High).
• Disable Asus GPU Tweak if you're using it.
• Disable MSI Center SDK if you're using it.
• If you're using Aorus Graphics Engine: disable autoscanning (auto OC).
• If you're using MSI afterburner, disable it before playing.
• If you're using GeForce Experience, disable its performance overlay.
• If you're using a Gen3 PCI riser with a Gen4 CPU, well don't.
• If your RAM is overclocked (XMP), set XMP off and see if it still crashes.
• Rename your Saved Games/DCS.../Config/options.lua to "options.lua.bak".
• If DCS is windowed, and NVIDIA G-Sync is "Fullscreen", set NVIDIA G-Sync to "Fullscreen and Windowed".
• Rename Saved Games/DCS.../Config/autoexec.cfg to "autoexec.cfg.bak".

[WWSmith]

It looks like you're using an AMD GPU. Which one exactly? Me and lots of other people on here with 5700XTs can't play DCS MT, and we've heard nothing about a fix since April. While the issue isn't unique to this 5700XT, there seems to be something about this GPU in particular which makes it especially incompatible with DCS MT. Like you, I have no problem running DCS ST.

Edited June 28, 2023 by WWSmith

[Paraso]

10 hours ago, Flappie said:

Hi. You're having the "DX device removed" symptom. It means DCS cannot reach GPU anymore at some point.

2023-06-09 03:37:51.945 INFO    EDTERRAINGRAPHICS41 (13092): surface5 gc() 5.327500 ms
2023-06-09 03:41:29.654 WARNING DX11BACKEND (13092): Can't map structured buffer.
2023-06-09 03:41:29.654 ERROR   DX11BACKEND (13092): Failed assert `0` at Projects\render\dx11backend_win8\Source\DX11StructuredBuffer.cpp:100
2023-06-09 03:41:29.656 ERROR   DX11BACKEND (13092): DX device removed. Reason: 0x887A0006
2023-06-09 03:41:29.657 WARNING DX11BACKEND (13092): Can't map structured buffer.
2023-06-09 03:41:29.657 ERROR   DX11BACKEND (13092): Failed assert `0` at Projects\render\dx11backend_win8\Source\DX11StructuredBuffer.cpp:100
2023-06-09 03:41:29.658 ERROR   DX11BACKEND (13092): DX device removed. Reason: 0x887A0006
2023-06-09 03:41:29.662 WARNING DX11BACKEND (13092): Can't map structured buffer.
2023-06-09 03:41:29.662 ERROR   DX11BACKEND (13092): Failed assert `0` at Projects\render\dx11backend_w

This can be caused by many things. Try these solutions:

• Ensure Windows power plan setting is correct (default or High).
• Disable Asus GPU Tweak if you're using it.
• Disable MSI Center SDK if you're using it.
• If you're using Aorus Graphics Engine: disable autoscanning (auto OC).
• If you're using MSI afterburner, disable it before playing.
• If you're using GeForce Experience, disable its performance overlay.
• If you're using a Gen3 PCI riser with a Gen4 CPU, well don't.
• If your RAM is overclocked (XMP), set XMP off and see if it still crashes.
• Rename your Saved Games/DCS.../Config/options.lua to "options.lua.bak".
• If DCS is windowed, and NVIDIA G-Sync is "Fullscreen", set NVIDIA G-Sync to "Fullscreen and Windowed".
• Rename Saved Games/DCS.../Config/autoexec.cfg to "autoexec.cfg.bak".

Thanks, I'll try the ones that apply to me and report back.

3 hours ago, WWSmith said:

It looks like you're using an AMD GPU. Which one exactly? Me and lots of other people on here with 5700XTs can't play DCS MT, and we've heard nothing about a fix since April. While the issue isn't unique to this 5700XT, there seems to be something about this GPU in particular which makes it especially incompatible with DCS MT. Like you, I have no problem running DCS ST.

 

I am using a 5700XT, I'll still try out some of the fixes but that really is a bummer.

[TonnyArno]

Well, really if it helps to anyone... I had the same issue but with Gigabyte RTX 3080 Ti Gaming OC. I tried something: changing RAM, removing XMP profile. I launched several stress tests like AIDA64 and OCCT. All stress test passed fine for a couple of hours with GPU and CPU don't reaching higher temp. than 73 C. But the game was crashing like in 10 minutes after entering the mission. I solved it by downclocking the GPU. The thing is - factory OC for my RTX 3080 Ti was 1710 MHz in boost. I downclocked it to 1665 MHz (which is a base clock for 3080 Ti in boost) and now it works perfectly. On your side - if all other hardware-consuming games don't crash for you, I don't think this is a hardware issue. Please, check whether you have some factory OC using GPU-Z. Clocks for your GPU should be 1605 MHz and 1905 MHz in boost. If it is higher - downclock your GPU using MSI Afterburner. Hope it helps.

[WWSmith]

TonnyArno, I tried your suggestion and found that my GPU was actually overclocked from the factory. I'd never considered this possibility before, having never tinkered around with hardware tuning, and just assumed that so long as I didn't overclock anything, then it wasn't overclocked. Anyway, I turned down the clock speed using MSI Afterburner. I've only just done this, and flew for about half an hour without a CTD, but I'll report back after more thorough testing.

[Paraso]

1 hour ago, WWSmith said:

TonnyArno, I tried your suggestion and found that my GPU was actually overclocked from the factory. I'd never considered this possibility before, having never tinkered around with hardware tuning, and just assumed that so long as I didn't overclock anything, then it wasn't overclocked. Anyway, I turned down the clock speed using MSI Afterburner. I've only just done this, and flew for about half an hour without a CTD, but I'll report back after more thorough testing.

Hi, how did you manage to confirm your GPU was overclocked from factory? Using GPU-Z my "GPU Clock" and "Default Clock" are the same value. Haven't checked MSI Afterburner yet. I have the "THICC III Ultra" 5700XT.
 

EDIT: Found that my GPU's P3 clock was set to 2084 mhz when the manufacturer defined stock boost clock setting was 2025 mhz. I set p3 to 2025 mhz using Radeon adrenalin software and attempted to run MT, CTD after a couple of minutes - log below. Manufacturer also lists base clock as 1810mhz and game clock as 1935mhz, not sure if I should be setting P3 clock to either of those, or what even counts as an OC in this case.

dcs.log-20230701-034601.zip

Edited July 1, 2023 by Paraso

[WWSmith]

Somewhat confusingly, MSI Afterburner only shows the boosted speed, whereas GPU-Z gives both the base and boosted speeds. In any case, both were above the manufacturer standard, so I used Afterburner to set it just below the recommended boost speed. I played about an hour last night and two hours this morning without a CTD yet.

Edited July 1, 2023 by WWSmith

[WWSmith]

No joy. I just had 2 CTDs in the past hour. I'll try reducing clock speed more, but I'm not hopeful.

Edit: the last CTD was not the DX Device Removed error. The log file from the one before that appears corrupt, because I can't open it, but the most recent was caused by a problem with an Apache texture, unless I'm reading this wrong. I've attached the log.

I'll continue testing with a different module.

dcs.log-20230701-140147.zip

Edited July 1, 2023 by WWSmith

[[DE] T-Bone]

I also have a 5700 XT and multiple CTDs, do I have the same error or is something else the problem with my log?

It is no longer possible for me to fly continuously, because the crashes sometimes appear after 2min, sometimes after 20. And all this started after the last update.

dcs.log-20230701-202425.zip

[WWSmith]

Strange that it only happened to you after the last update. MT hasn't worked for me since either the initial release or the very first update, I can't remember now. This morning I thought that underclocking might have fixed it, but then I had another DX Device Removed crash this afternoon. Unless I just need to underclock more, I'm all out of ideas.

[TonnyArno]

@WWSmith, I noticed that after reboot I need to activate my underclocked profile again in Afterburner. Also, the underclocking is reset after drivers update. BTW, even after it, sometimes in game I can notice sort of a black screen appeared for a split second with a micro-freeze, but game just doesn't crash. Anyway, in my opinion this is a dirty hack. I mean, let say, all your other games dont crash, and system stability tests pass fine, so no reason to think that the problem is with your hardware. This "device removed" problem shows that your GPU was physically removed or your driver stopped working. But believe me - there is a bunch of ways to drop the driver from the application. I had this issue writing a simple program in CUDA. You can drop a driver or receive a BSOD by writing to non-allocated memory of your GPU (in my case it was so-called unified memory), and such problem may appear on one GPU and be completely hidden on another. So, to summarize - developers should debug it and find a problem in a source code. First of all - they should eliminate the messages that we see in log file that some model wasn't found or some texture is missing. I guess this is not normal in a campaign which was released already some time ago and should work smoothly.

[WWSmith]

@TonnyArno I just had another CTD, and I ran GPU-Z immediately afterwards to confirm that I was underclocked. I had turned the base clock speed all the way down to 1350 from 1605, while boost was at 1895 vs the default of 1905. (Incidentally, how can you tune base and boost independently?) Anyway, as I've mentioned before, I can run DCS single-thread all day long with no crashes, regardless of underclocking/overclocking, so there is pretty clearly some issue with multithreading. I really wish we could get some official word on this, even if just to say that they still can't figure it out, because it's now been about 3 months since I've heard anything at all.