[1.1.21 (build 58049)] DXGI_ERROR_DEVICE_RESET started this week

[jodavaho]

Every 15-20 minutes, screen goes dark, pc resets, bios post failed, then resets once more, and back to windows just fine.

So far, no issues with other games / long running programs.

This issue started this week. I've had about 300 hours of playtime to date with zero crashes prior to this. My system is new-ish, as of late december, but I've had many full Factorio days without issue. The recency makes me suspect some new situation this week.

My logs, mods (none), and save are attached.

[jodavaho]

I also have a windows stack dump, and memory map, but don't want to post those for various (perhaps obvious) reasons yet.

[jodavaho]

Tried force-opengl. Same crash. This time I heard the windows hardware disconnect sound after screen went dark and factorio game sounds ended. Clearly not getting video to my monitor.

More information:

GTX 770
NVIDIA driver: 461.40, release 1/26/2021

OS Name Microsoft Windows 10 Pro Insider Preview
Version 10.0.21286 Build 21286
System Model NUC9i9QNB
System Type x64-based PC
System SKU BKNUC9i9QNB
Processor Intel(R) Core(TM) i9-9980HK CPU @ 2.40GHz, 2400 Mhz, 8 Core(s), 16 Logical Processor(s)
BIOS Version/Date Intel Corp. QXCFL579.0054.2020.0810.2239, 8/10/2020
SMBIOS Version 3.3
Embedded Controller Version 24.36
BIOS Mode UEFI
BaseBoard Manufacturer Intel Corporation
BaseBoard Product NUC9i9QNB
BaseBoard Version K49243-403
Platform Role Desktop
Secure Boot State On
Installed Physical Memory (RAM) 64.0 GB
Total Physical Memory 63.9 GB
Available Physical Memory 56.6 GB
Total Virtual Memory 95.9 GB
Available Virtual Memory 87.0 GB
Page File Space 32.0 GB

[bormand]

So far, no issues with other games / long running programs.

windows hardware disconnect sound after screen went dark

Can it survive an hour or so in something like 3d-mark stress-test?

Why I ask this? Because the system is very powerful and compact at the same time. Cooling and power supply may be an issue.

[jodavaho]

> Can it survive an hour or so in something like 3d-mark stress-test?

I haven't tried, but I've done this:

- Played various 3d-intensive games for 3+ hours just about every day since December (thanks pandemic, I guess)
- Benchmarked CPU/Mem using intel tools
- Played factorio for 12+ hours a few days in the last few months without issue (thanks pandemic again!)
- Checked system temps regularly before / after shutdowns -- but not GPU temps, so perhaps

The sudden start (last few days), and predictability (10-15 m between crashes) makes this seem unrelated to my overall system.

I'll try it out.

[jodavaho]

3D Mark has been running for 15-20 minutes, with a max GPU temp of 79C. I''ll run for another 30mins.

[Rseding91]

Considering we haven't changed anything around graphics in quite some time (more than a year as far as I know) and you say this started happening randomly; that says failing hardware to me.

Especially since this comment:

This time I heard the windows hardware disconnect sound after screen went dark and Factorio game sounds ended.

That says windows detected the hardware had stopped responding and restarted the graphics driver.

I had my GTX 970 die on me about 7 months ago after years of working without issue; it just started having weird issues/artifacts in some games and caused overall system instability. Never BSOD-ed on me though. Replacing it fixed all of the issues immediately.

So, it's just up to you to figure out what hardware is failing.

[jodavaho]

Yeah, you're probably right on this. I don't know why Factorio was struggling so badly, but I did all kinds of stress tests to no avail (obv not thermal issues, for example), opened the case, re ran wires, re-seated the video card, and rebooted. Now it's as solid as ever, so I think it was a sudden electrical issue w/ the card.

A few other minor things:

1. Disconnected a USB hub and attached all peripherals directly
2. Saved to a new game slot.

Just for archival sake, I don't think those had anything to do with it.

[cbhj1]

Former computer repair here, never underestimate a reboot or component reseating.