DXGI_ERROR_DEVICE_HUNG error: Is there a solution?

[Pete Dowson 558]

Well, after a couple of days without the dreaded DXGI device hung error, I just got one -- immediately when going to EGLL (my heaviest loading airport). It didn't even get time to display!

This is with the less used GPU disabled. So I don't think that video card is faulty.

I've no idea what is going on, nor what to try next. These errors are all recent, infrequently with P3D4.3 but now a real stopper with 4.4. I've had this system nearly 18 months now, so something else software related is wrong.

I'll try using DDU to clean up this video driver, and install the latest.

Pete

 

[Guest]

15 hours ago, cj-ibbotson said:

The retail version of the 9900K is now £525 on Overclockers where I believe you bought yours so I might just go the whole hog and change as much as I can now, might just order a new PSU as well.  Ive held back as concerned of heat of the 9900K, currently have a Corsair H100i v2 which is 240mm long.  Might buy a 360mm unit H150 i think and a new case

I was able to secure another 9900K local (US) by doing the "combo" purchase on NewEgg ... actually comes out pretty cheap ... this is for another build I'm doing for a development machine ... so it appears that 9900K demand is perhaps dropping a little ... here's the fancy packaging for retail (seems a waste of plastic):

 

But I would agree that maybe you should do a complete new build and make sure everything is good on the wiring side.

The path to diagnostics is to always keep an open mind.  There is a very very small possibility that P3D could have a software bug that exposes hardware weakness, but that would be extremely low on my priority list relative to the many other possible causes for DEVICE HUNG.  Sounds like you are making progress, anyone suffering DEVICE HUNG can definitely learn from this thread.

Cheers, Rob.

[Guest]

14 minutes ago, Pete Dowson said:

I'll try using DDU to clean up this video driver, and install the latest.

That's a good idea, but you might also try to do some data logging, verify card is seated, verify power cables ... per Bob, I HIGHLY recommend you get one of these PSU testers or the Thermaltake unit.

Cheers, Rob.

[cj-ibbotson 77]

18 minutes ago, Rob Ainscough said:

I was able to secure another 9900K local (US) by doing the "combo" purchase on NewEgg ... actually comes out pretty cheap ... this is for another build I'm doing for a development machine ... so it appears that 9900K demand is perhaps dropping a little ... here's the fancy packaging for retail (seems a waste of plastic):

 

But I would agree that maybe you should do a complete new build and make sure everything is good on the wiring side.

The path to diagnostics is to always keep an open mind.  There is a very very small possibility that P3D could have a software bug that exposes hardware weakness, but that would be extremely low on my priority list relative to the many other possible causes for DEVICE HUNG.  Sounds like you are making progress, anyone suffering DEVICE HUNG can definitely learn from this thread.

Cheers, Rob.

Looks good Rob.  Im still to order as do not have time to research everyone I need to buy and so busy with work.  My main concern with a 9900K is the heat.  Ive experienced heat issues the past week with my 6700K almost hitting 100 degrees with P3D running and HT is even off and only a 4.5 overclock at 1.3v  Ive reconfigured fans in their push pull setups etc and changed settings in the BIOS for the case fan and also using Corsair Link to make the pump and fans on the H100i V2 AIO cooler run better and temps are back to normal.  Scared going back there with a 9900K as P3D adds a lot of heat.  Id be looking at buying a 360mm Corsair unit, the H150i I think but wonder if its still sufficient.

Ive anaged to access my system now as the psu is working again and here is a link to the GPU-Z log file generated yesterday.  The sim had been running for 14 hours altogether when the crash occured though logging I had only started an hour before the crash which happened at 14:32, you can see GPU Load shows 2 entries at 100% which is totally out of sync with all before it.  Sim wasnt even flying it was in slew mode and set to around 20000ft and 160Kt speed, was daytime too and just moving over the eastern Alps though the error can happen anywhere.  It was happening more at dusk / night / dawn in the past but now daytime.  I only got 14 hrs by lowering the gpu Power Limit to 75%, at 80% it crashed in a few hours.  This is why I really wonder when people say 'doing xx fixed my sim' given it can take an extremely long time for the error to show.

https://www.dropbox.com/s/c6ris4qjfxjzdpc/GPU-Z Sensor Log.txt?dl=0

Chris

[Pete Dowson 558]

24 minutes ago, Rob Ainscough said:

but you might also try to do some data logging

How, please? What do I need to use? GPU-Z or Afterburner?

24 minutes ago, Rob Ainscough said:

verify card is seated, verify power cables

Did that when looking at removing the middle card of the original 3-way SLI arrangement. It wasn't so easy to do that. Cooling cables all over the place, linking each card to a separate fine on the perimeters of the case. To get the card out I'd have to do a lot more than I anticipated, so all i did was remove the now defunct SLI links and ensure the card was disabled in Device Manager.

Pete

 

Edited December 7, 2018 by Pete Dowson

[pgde 571]

42 minutes ago, Pete Dowson said:

How, please? What do I need to use? GPU-Z or Afterburner?

Did that when looking at removing the middle card of the original 3-way SLI arrangement. It wasn't so easy to do that. Cooling cables all over the place, linking each card to a separate fine on the perimeters of the case. To get the card out I'd have to do a lot more than I anticipated, so all i did was remove the now defunct SLI links and ensure the card was disabled in Device Manager.

Pete

 

Pete -- use GPU-Z and check the log option on the lower left of the GPU-Z screen. It is well worth the time!

Happy Holidays!

Peter (the other Peter)

[Jim Young 2,356]

1 hour ago, Pete Dowson said:

immediately when going to EGLL (my heaviest loading airport). It didn't even get time to display!

As I recall, others were having this issue with EGLL and Gatwick.  Possibly a problem with the UK scenery in P3DV4 (not sure what you have installed).  I'm sure that thousands of members already own UK2000 if that is the scenery you have installed and not all of them are having the dxgi error.  I am tempted to buy it just to test it.  Does the error still occur with all add-on scenery removed? (Go to Documents/Prepar3D Add-ons folder and move all of the contents to a temporary directory; only default scenery will load).  There could be just an object or autogen that is not compatible.

[Bob Scott 7,715]

1 hour ago, cj-ibbotson said:

Looks good Rob.  Im still to order as do not have time to research everyone I need to buy and so busy with work.  My main concern with a 9900K is the heat.  Ive experienced heat issues the past week with my 6700K almost hitting 100 degrees with P3D running and HT is even off and only a 4.5 overclock at 1.3v  Ive reconfigured fans in their push pull setups etc and changed settings in the BIOS for the case fan and also using Corsair Link to make the pump and fans on the H100i V2 AIO cooler run better and temps are back to normal.  Scared going back there with a 9900K as P3D adds a lot of heat.  Id be looking at buying a 360mm Corsair unit, the H150i I think but wonder if its still sufficient.

Ive anaged to access my system now as the psu is working again and here is a link to the GPU-Z log file generated yesterday.  The sim had been running for 14 hours altogether when the crash occured though logging I had only started an hour before the crash which happened at 14:32, you can see GPU Load shows 2 entries at 100% which is totally out of sync with all before it.  Sim wasnt even flying it was in slew mode and set to around 20000ft and 160Kt speed, was daytime too and just moving over the eastern Alps though the error can happen anywhere.  It was happening more at dusk / night / dawn in the past but now daytime.  I only got 14 hrs by lowering the gpu Power Limit to 75%, at 80% it crashed in a few hours.  This is why I really wonder when people say 'doing xx fixed my sim' given it can take an extremely long time for the error to show.

https://www.dropbox.com/s/c6ris4qjfxjzdpc/GPU-Z Sensor Log.txt?dl=0

100 deg at 4.5/1.3v is not normal...perhaps your loose power connector was causing the H100i pump to run slow or intermittently?

Your log shows that, immediately preceding the DXGI error, both the memory controller and PCIe controller loads on the GPU drop from a fairly steady 22/3% to 0/0, at the same time GPU load spikes to 100%.  I don't see a temp or voltage problem...it looks like either a driver problem or graphics hardware.  The immediate dropout or failure to start the OCCT GPU test is also troublesome.

Regards

[Pete Dowson 558]

35 minutes ago, pgde said:

Pete -- use GPU-Z and check the log option on the lower left of the GPU-Z screen. It is well worth the time!

Okay.  Not today though. Finished for the day.

I cleaned up the video driver stuff with DDU and installed 471.22. Oddly, after that I couldn't set the refresh rate on 2 out of 3 of my projectors to 30 Hz. Below 60 it only offered 50, 25 and lower -- but only on those two (all three are identical Acer projectors)!

So re-booted (again), Then I was offered 29 on those, no 30. so now all three are set to 29Hz instead of 30 as before! Weird.

Anyway, in exactly the same scenario as before -- same place, same time, it was okay.

I'm not (yet) assuming the driver tidy & update helped here, so next time I go into one of these high demand scenerios, I'll first enable GPUZ logging.

Thanks,
Pete

 

[cj-ibbotson 77]

59 minutes ago, w6kd said:

100 deg at 4.5/1.3v is not normal...perhaps your loose power connector was causing the H100i pump to run slow or intermittently?

Your log shows that, immediately preceding the DXGI error, both the memory controller and PCIe controller loads on the GPU drop from a fairly steady 22/3% to 0/0, at the same time GPU load spikes to 100%.  I don't see a temp or voltage problem...it looks like either a driver problem or graphics hardware.  The immediate dropout or failure to start the OCCT GPU test is also troublesome.

Regards

I will reset my CPU tonight to stock speeds and try OCCT GPU test again just to rule out any issues with click speeds. It runs Furmark and other stuff no problem. Can you think of any other stress tests for GPU?

EDIT

Booted the pc with an XMP profile so memory and CPU etc back at stock values and OCCT stopped in immediately when I tried the GPU test

Chris

Edited December 7, 2018 by cj-ibbotson

[Cruachan 705]

Hi,

How about this for coming out of left field:  Mention has been made about checking to ensure the GPU has been properly seated, but nothing has been said about the size and weight of these cards. They are heavy! Despite being fixed to the case with 2 screws and the retaining latch at the end of the PCIe slot, the card still sags slightly under it’s own weight. This must be applying an unequal stress along the length of the slot. Is it not conceivable that potentially this might lead to an unreliable electrical connection which could be affected by the heat induced expansion and contraction of the slot connectors? Not sure how effective these latches are in terms of guaranteeing electrical integrity, but the assumption is that the MoBo manufactures have confidence in this reference design despite the size and weight of modern GPUs.

A while ago I upgraded from an SLI setup with 2x980Ti to a single 1080Ti. Removing the upper card proved tricky and I managed to break the retaining latch. The MoBo had proved its reliability beyond any doubt and I was reluctant to change it just for the sake of one broken latch. However, mounting the card under these unsatisfactory conditions was not really a sensible option, although could be tried despite the near certainty that, over time, the rear end of the card would slowly lift out of the slot. So what to do? Head scratching time!

The answer came in the form of 2.5 x 100mm Nylon Cable Ties! These were/are assembled in a daisy chain. One end is tethered to a suitable point near the top of the card and just below the card’s power connectors; the other is attached to the back of one of the drive bays of the case. This cable tie assembly is angled upwards and backwards towards the MoBo such that, when tightened gently, it seats the card firmly in its slot, elevates it slightly to maintain a stable horizontal configuration and, as such, ensures secure and even electrical contact of all the slot connectors with those on the card. After many months using this solution with the 1080Ti I have experienced no GPU related issues.

Might be worth a try even if your retaining latch is still intact. 

Regards,

Mike

 

[Bob Scott 7,715]

I use a wooden dowel cut to length to support the top end (opposite side from card slot) and take the side load off the PCIe slot(s) holding the GPU(s).

Some of these 1.5-2x width graphics cards, equipped with oversize heat pipes, heat exchangers, and triple fans/fan shrouds can get pretty heavy.  I have never been comfortable with that much torque being applied to the card slot.

Regards

[cj-ibbotson 77]

I recently reseated my card when I had to take it out to test my sons 1080.  Ive also tested my own with the case on its side on the floor so there is no sag.  The power cables are also installed coming through the back panel of the case to the top of the cards so that the cables effectly hold the end of the card up to reduce the risk of it sagging

Chris

[cj-ibbotson 77]

On 12/6/2018 at 9:44 PM, Rob Ainscough said:

This is normal after a full power down (i.e. PSU unplugged or switch in Off position), did you wait a few seconds more, it should automatically attempt to power on again.

If you suspect the PSU and can get the PC to fire back up, if you run Time Spy (3D Mark) it will most definitely stress out your CPU and GPU (well beyond anything P3D can do) ... during my many Time Spy tests it would draw so much power that the APC UPS/line conditioner that I plug my PC into, would trigger and the APC's fan would start up as it attempts to compensate for power draw during the GPU tests.

Time Spy is a beast of a testing/performance software, I could pass every other test I tossed at it, but Time Spy would uncover stability issues that all other tests wouldn't.

Cheers, Rob. 

Rob Ive just got round to running Time Spy.  I forgot I got a copy of 3D Mark Advanced with my 1080ti so I installed it, updated it then ran both the Extreme Benchmark and the Extreme Stress test for 20 loops.  Before this I reset my gpu power limit setting back to 100%..a setting which causes DXGI crashes in P3D.  Benchmark test passed no problem and so did the extreme stress test with a passmark of 97.5%

These sorts of things just leave me wondering again if my card is 'faulty'.  During stress test the max gpu temp was 83 degrees. max cpu was 70 degrees, currently running it at 4.7 (HT off) 1.3v

Here is the log file of the Time Spy stress test and all looks fine on a quick glimpse, core clocks etc dropped briefly near the start which was probably when the software was loading

https://www.dropbox.com/s/tv8jk42zgc30bda/GPU-Z Sensor TIME SPY Log.txt?dl=0

Chris

[Snuffleupagus 327]

3 hours ago, Cruachan said:

Hi,

How about this for coming out of left field:  Mention has been made about checking to ensure the GPU has been properly seated, but nothing has been said about the size and weight of these cards. They are heavy! Despite being fixed to the case with 2 screws and the retaining latch at the end of the PCIe slot, the card still sags slightly under it’s own weight. This must be applying an unequal stress along the length of the slot. Is it not conceivable that potentially this might lead to an unreliable electrical connection which could be affected by the heat induced expansion and contraction of the slot connectors? Not sure how effective these latches are in terms of guaranteeing electrical integrity, but the assumption is that the MoBo manufactures have confidence in this reference design despite the size and weight of modern GPUs.

A while ago I upgraded from an SLI setup with 2x980Ti to a single 1080Ti. Removing the upper card proved tricky and I managed to break the retaining latch. The MoBo had proved its reliability beyond any doubt and I was reluctant to change it just for the sake of one broken latch. However, mounting the card under these unsatisfactory conditions was not really a sensible option, although could be tried despite the near certainty that, over time, the rear end of the card would slowly lift out of the slot. So what to do? Head scratching time!

The answer came in the form of 2.5 x 100mm Nylon Cable Ties! These were/are assembled in a daisy chain. One end is tethered to a suitable point near the top of the card and just below the card’s power connectors; the other is attached to the back of one of the drive bays of the case. This cable tie assembly is angled upwards and backwards towards the MoBo such that, when tightened gently, it seats the card firmly in its slot, elevates it slightly to maintain a stable horizontal configuration and, as such, ensures secure and even electrical contact of all the slot connectors with those on the card. After many months using this solution with the 1080Ti I have experienced no GPU related issues.

Might be worth a try even if your retaining latch is still intact. 

Regards,

Mike

 

I think this may be my solution and is something I noticed too when I reseated the card. It’s a heavy card so I used extra props to hold it up. So far it’s been working. I find it kind of odd that it just started to begin the day I did the 4.4 update and not having moved my tower in months. But hoping it’s a possible solution!

Edited December 7, 2018 by pvupilot