So I'm trying to render the Spring demo but everytime I render it using GPU it crashes but if I render it on my CPU it doesn't crash. Does anyone having similar issues? My computer spec are 32gbs of ram, ryzen 2600x and gtx 1060. I'm thinking blender still has some stable issues still.
While I can't pin-point the source of the issue, it seems to be caused by a problematic interaction of modal operators by some add-on or script and the auto-save feature which causes the TDR-Delay (Timeout Detection Recovery) to be exceeded.
The current workaround is to restore the factory settings. This can be done in Edit > Preferences. First load the factory settings by clicking on the button in the bottom right corner. Then save the preferences to make these changes permanent.
I just started using a Dell G7 with Nvidia graphics, and I immediately started having the dreaded "Your computer had a problem ... we'll restart it for you ..." with a VIDEO TDR FAILURE error message. My complete setup is as follows:
System: OS Name Microsoft Windows 10 Home Version 10.0.17763 Build 17763 System Manufacturer Dell Inc. System Model G7 7790 System Type x64-based PC Processor Intel(R) Core(TM) i7-9750H CPU @ 2.60GHz, 2592 Mhz, 6 Core(s), 12 Logical Processor(s) BIOS Version/Date Dell Inc. 1.6.0, 6/1/2019
Video: Adapter Description NVIDIA GeForce GTX 1660 Ti Adapter RAM (1,048,576) bytes Driver Version 220.127.116.1153
Blender: version: 2.80 (sub 75), branch: master, commit date: 2019-07-29 14:47 Rendering with Cycles or Eevee. Graphics interface: CUDA (unable to find version)
So, being a software veteran of many, many years, I'm tolerant and inquisitive. I found some reference (and subsequently lost it) to a Registry setting for TDR Timeout. It seems that the graphics system interface keeps tabs on whether the graphics driver is hung or not, and if the interface hangs for more than a certain time, it throws a fit and dumps some data somewhere. It then proceeds to restart the computer.
I looked around in the registry and found the following:
The initial or default setting on my system (as described above) was "Value not set". Well, that strikes me as terribly uninformative, so I had read somewhere that the default was meant to be two seconds. I upped it to first 5 and then 10 seconds, restarted the computer and Blender, and proceeded to try to make Blender and the video fail.
Interesting observation: when it seemed that Blender was totally unresponsive, I started counting seconds until the crash. When I had the timeout set to 5 seconds, the crash came 5 seconds after I noticed the Blender stasis. When I moved the timeout to 10 seconds, I could count 10 seconds before the apocalypse. I also noted that several times when Blender became unresponsive, it might recover and become responsive again as long as the delay was less than the 10 seconds I had set.
What this says to me is that the Nvidia driver coupled with Blender 2.8 somehow generates an occasional spate of "hung". The pair will sometimes recover, and all is fine. These hang times varied from less than a second to several, but fewer and fewer times did the hang exceed 10 seconds.
DISCLAIMER: I messed with my Registry on my own computer. Do not do this unless you are a trained Windows technician; you could damage or destroy your operating system or hardware. Don't say I didn't warn you!
Update 9/19/19: Nvidia came out with a "studio" driver update just last week. The studio version is meant for "production users" as opposed to gamers, so I downloaded and installed. It's been a week now with no die-and-reboot. Things are looking up. I've been switching from Blender to a browser (Firefox) to GIMP to Camtasia with no problem. It's only been a week, but I used to get these every day before. Maybe the Nvidia driver was it. I'll wait a week and then reset my TDRWATCH timeout to default and see what happens. One thing I did notice is that a really heavy graphics use will sometimes "pause" for one to 5 seconds and then resume. Here's hoping.