GPU stops responding, win10, spec.material with fake shadows

[deadakos]

I've upgraded to win10 and after a longer period, I restarted to work with octane on a personal project.
Everything worked like a charm until I started to render a multiwall material.
Until the material is not specular with fake shadows enabled, there is no problem at all. When I change it to specular with fake shadows enabled, my gpus starts not responding and I receive messages from the OS that "Display driver stopped responding and has recovered Display driver NVIDIA Windows Kernel Mode Driver". The driver version is 355.60
They stops one after the other, not all at the same time.
I have 1 GTX 580 (not used for rendering); 2 GTX 590 (1 MSI and 1 GAINWARD)

I modeled in Sketchup, the obj was exported with the old exporter. Formerly I already made such materials without any problem. I tried previous Octane versions but the same happens there too. I tried older objs again (where I know that it worked) and the same happens.
I tried to open the objs with other softwares, but I can't see any problem with them.

It makes no difference if I use PT, PMC or DL kernels.

Would you please help me with this problem?

[haze]

Please do double check that you're not rendering on your display device.

Some questions:

- So if you use a specular material but have fake shadows disabled, there's no problem?
- How big is the scene? What is your VRAM usage like?
- Are you able to reproduce this with a simpler scene?
- Does the same issue happen when rendering only on one device?

A last ditch effort would be to try a different driver version.

[face_off]

If you install Octane Standalone from the .exe installer (not the .zip) it will set your Nivida driver time-outs correctly, which will potentially resolve this issue.

Paul

[deadakos]

I double checked everything, and the situation is more than mystic.

My answers for first:
@Paul: downloaded and reinstalled Octane 2.24 with the installer, not from the zip archive and nothing changed unfortunately.
I found a Microsoft forum thread yesterday, where one were suggested to add a registry entry of TdrDelay, with the value of 8 which used to help in case of slow GPU answers. I think you thought about this when you wrote "Nvidia driver time-outs", hopefully.

@mist:
Double checked and I didn't allow my display device to work as a renderer GPU.
- Yes, if the "fake shadow" property of the specular material is disabled, everything is OK.
- VRAM usage is 35,3 MB used from the 1348 MB available.
- No, I can't reproduce it with a simpler scene.
- Yes, it makes no difference if I enable just one GPU or, both GPU of one 590.

This morning I used DDU (Display Driver Uninstaller) to completely remove Nvidia drivers, and downloaded the lastest WHQL driver (which is the same which was installed previously (355.60)) and installed it. Nothing changed.

I finally started to to re-model this part of my scene, and I tested the result step-by-step. I even started to scatter the repeating part to minimize the error. For first I couldn't reproduce the error so I thought that it was just a modeling error, but as I proceeded I started to receive the error message again.
And now comes the interesting part!

After a restart of the completely freezed computer caused by this error, sometimes the error happens again, but more often don't, even with the old model!

Just to be sure, my PSU is a Thermaltake Toughpower 1350W.

Any idea?

[deadakos]

It seems that the clean reinstall (not the upgrade from win7) and setting TrDelay to 0 in the registry solved the problem.
http://answers.microsoft.com/en-us/wind ... 3c6?auth=1

Cheers
Akos