Doesnt seem to be an issue with the new dependency graph.
[run blender with --enable-new-depsgraph option]

Not many people will experience this bug. Maybe automaticly disabling avx2 in this case?

@Lukas Stockner (lukasstockner97) , @Brecht Van Lommel (brecht) feels like a codegen bug, in a compiler we're soon gonna drop anyhow, not sure if it's worth spending time tracking it down.

I tried the vc14 version and it works. So this bug seems to be only happening when using Avx2, Windows and the old compiler. Very exclusive little bug. :-)

I can only seem to reproduce this with msvc2013 and avx2, @Falk Müller (xharthok) can you try a build from the bot marked with 'New Compiler' (has vc14 in the filename) and see if it works for you without having disable avx2?

Thanks for the tip! I disabled AVX2 and everything renders fine.

Thank you Philipp :)

Confirmed by @LazyDodo (LazyDodo), only seems to happen on AVX2-capable processors on Windows.

After further research I found this article: Preventing Cycles Crash after CUDA timeout

Ah, view order seems to have an effect (if left 3D view is in local > no problem, if right view is in local > problem)
Will have a closer look...

I have just downloaded the latest build and problem is still there:

Hm, weird, still cant reproduce (2.79, 2.79b and master on linux, 970m, 390.59 drivers) -- rendering to UV Image Editor also fine here...

Hi. it seems that this situation occurs only when we render scene in the same window, but when we render to a new window everything looks good:

I started experiencing such an issue after I updated one week ago my Win10 Pro 64bit OS from Build 1511 to 1803 on an ASUS G752VS notebook (NVIDIA GeForce 1070). Reinstalling the graphics drivers (398.11) with a clean install didn't fix the issue. The error appears randomly when switching the 3D View to Rendered mode; closing and restarting blender and re-opening the same .blend project file usually allows the GPU to render without errors.
The Blender version is 2.79b 64bit official portable release; the same issue appears on the latest buildbot releases 2.79.4.
The following are screenshots of the issue appearing after switching the 3D View to Rendered mode inboth Blender 2.79b and 2.79.4:

Can confirm the behavior, not sure though if there is a reasonable fix.

Awesome News Brecht!!!!!

Oh, @Brecht Van Lommel (brecht) was quicker :) [phab hung here for a bit, so my comment came late...]

Can confirm this,

Apparently, it is enabled already. So closing as resolved.

Uninstalling „Nvidia Experience“ does not fix the problem to me.
Cuda Error in line 1712 still occurs in complex scenes.
I `m using a Zotac GTX1080TI, Windows 7, latest Nvidia driver and all Blender versions (nightly builds, RC and Release versions. Modifying the reg keys does not help also.
Im sure that I did never had this issue with my older graphic card GTX 1080 (without TI)
It seems that some code is not able to get or copy/clean the content into/from the vram which is higher than 8GB (for me the main different to the older card).

Hey all,

Glad to hear! Closing then...

I disabled sandbox mode on Comodo and gave it a bake in the June 15 version of blender using my GPU and it worked as it should. Problem solved. Thank you for all the assistance.

Thanks for the suggestion. We don't track feature requests in this tracker, http://rightclickselect.com/ is the right place for that.

You could just drag and drop the blender.exe into the command promt and add --debug-cycles to the end

Just wanted to drop in and say good job @L. E. Segovia (amyspark) !

My GPU does only have 1 Gb of VRAM, so that could be the case.

Lukas thank you so much for taking the time to reviewing this and provide me with that feedback!

Seems like GPU only then.
Could it be that you are running out of GPU memory?
(if I look at my NVIDIA Server Settings I have it peaking at a little less than 2 GIGs -- which might be a little too much for the GTX 650 Ti...)

Fixed in rB7bf40236893c.

Thanks for the patch! I'll go ahead and commit it.

patch at https://developer.blender.org/D3472

Try setting the lamp size to 0 or a small number, in this .blend it is set to cover a big part of the hemisphere. We have a to do item to clarify this by changing the sun size to display as an angle.

The source of the problem is the use of ray_offset() to prevent self-intersections for rays that originate from a transparent surface. A solution to this problem is outlined in the paper "Robust Iterative Find-Next-Hit Ray Traversal": http://www.sci.utah.edu/~wald/Publications/2018/nexthit-pgv18.pdf

Please. revert them to the original version! because without clearcoat gloss you can't make cool materials like jelly, plastic stuff like that!

More than a week passed without a reply.

The system has CUDA-9.2 installed. Seems it's not compatible yet. Compiled with CUDA-8.0. Everything works fine.

After upgrading to 10.13.5 with latest Nvidia drivers the problem persists. Any idea if I could do anything else? Would 2.8 use a newer version? I guess EEVEE would be even better anyway for that type of shadings.

I just checked the project using another Windows drive that I had around and it seems to be working fine. So I guess the problem is as you said with the OpenGL in my OSX version or something. Will update my system as there is a new one and see if the new drivers solve it. Any idea about what I could do in the case that would not solve it?

I just checked the file using another iMac and the Matcap looks correctly in Material shading. OSX 10.10.5 with Ati Radeon HD 2400 XT. I was wondering if this has to do with Nvidia drivers as you said. Will check the other person graphics card brand as he reported not working on Windows.

The project as stated was tested by another user using a PC on windows and he had the same result while he could have the right result using a Macintosh. I was hoping for you or someone else to download the file and test if you could replicate the problem. My graphics card is up to date and I never had any problem visualizing everything as the way things should look like.

Its highly likely that this bug is specific to your hardware/driver configuration. These kinds of display issues are often caused by OpenGL driver bugs.

Okay, sure. Blend file and info OS/Hardware find above. To be precise about versions:

Please use the new bug report template and provide all the requested information:
https://developer.blender.org/maniphest/task/edit/form/1/

Can confirm this in current buildbot builds (whereas 2.79b seems fine).
Having trouble building with OSL atm. (due to llvm build issues - but working on that to find the cause...)

Hey all,

With threshold set to 0.0f (unfortunately hard-coded right now), memory usage of the color texture in this example is reduced by ~33%. If set to 0.2f, memory usage is reduced by ~66%, with no significant visual loss.

Until this is fixed, there is a workaround:

Confirming, but not sure if this can be solved for the Principled BSDF (as it is kind of a mix between different BSDFs [and thus raytypes?]).
Described behavior is also true for a simple 50/50 mix between a standard diffuse and a standard glossy.
So in this situation glossy contributes... (if you set glossy bounces to zero then contribution in Diffuse Indirect pass also vanishes...)

Having the same issue as everyone else- the denoising filter turns all tiles black. Deactivated it and everything rendered fine.