The File: Bliblubli posted renders without error on Build: 57f937ad5a Blender 2.78.4 on an RX 480. Must be something that has changed after. Ill keep digging.

To be clear, this should be committed after 2.79 is released since we are in bugfix only mode.

I don't immediately have a good suggestion to optimize this. It's possible in principled to use SIMD for the cross product, normalize and fast_acos (which is likely the slowest part), but that's not so simple to implement. To me performance seems acceptable.

then problem in 3DviewPort, (grid and unit we see), Because blender changes the units to expand the range, and Improve quality. PS. The difference between imperial and metric is not taken into account

Ok, then assume they changed to 10^-4 W.

but not in square

Yes, but you are not taking into account that the units of light emission changed.

beacose Cycles works winth Blender_Unit (I thing)

But, for your result (without scaling)

in thouse example plane and point has been scaled. Please pay attention at the grid size.

I'm not sure what you did in that test, but here I don't see any differences in the render results when changing units? Blender units only affect how the value is displayed in the UI, not the actual render.

Ok, this is more about the unit system in general then like the discussion in D1799.

I meant that Cycles is based on real dimensions in real units, but in the blender users can change units of scene in the scene (meters centimeters, and so on), and when the time comes to render, the scaled scene renders inaccurately as the light fades out not linearly My opinion is to enter in the Cycles the scaling factor of the scale or to make clarity with the scale units in Blender.
When you select units in cm, then simulate and then change to meters, then the model will increase by 100 times, but it is only necessary to change the prefix from cm to m and change the number that is displayed on the screen and not scale the model relative to the scene.
In my opinion, the bug is in Blender and not in Cycles, but Blender needs to change everything from the center, and therefore, it is necessary to Check the scale-dependent scene modules based on Blender, therefore it is possible to enter the coeficient in Cycles temporarily to solve this problem, before solving it in Blender.

Ok, good to hear you found a solution, sounds like there is not much we can do about this in Blender.

After playing around with this setup now for a day I just found this link
https://egpu.io/forums/implementation-guides/dell-precision-7510-akitio-node-galax-gtx-1080-hof-its-working/
where the user disabled the internal nvidia GPU and then the external GPU worked in CUDA-Z (!)
... and now it works in blender too!
(not a perfect solution but at least I didnt buy the Akitio +1080Ti for nothing - and maybe nvidia can fix this in the future)
PS: THX brecht for your help!

Yes, drivers are from nvidia. Control panel shows version 384.94
BUT Cuda-Z says "no compatible CUDA devices found" when the external gpu is plugged in :-(
... but nvidia gpu monitor shows eg Chrome running on the GPU
any ideas?

Also, did you install the drivers from nvidia.com? What is the exact driver version?

Two things to check:

• Does CUDA work in another application like http://cuda-z.sourceforge.net/ ?
• Run Blender from the command line with --debug-cycles and attach the output here.

This is my first multi-line patch. I'm happy to see it in master!

In T51408#432903, @Alexander Gavrilov (angavrilov) wrote:

Well, I think that emission and cutout transparency (i.e. holes) can be handled perfectly well by mixing in other nodes, unlike translucency (diffuse transmission, which should share energy with diffuse after glossy), so it's not a priority; and the node may have enough parameters already. Also, is somebody writing a manual entry for it already? ;)

I can't reproduce this with the hardware I have available (no macOS here). Maybe @Lukas Stockner (lukasstockner97) can take a look?

Looks good to me, thanks for the patch!

I tried to repro this with Intel + macOS (with a simplified megakernel since the full one doesn't work), but denoising seemed to work ok. @Mai Lavelle (maiself), do you have a way to test this perhaps? Not sure who maintains macOS OpenCL now.

Normally Cycles render results are not affected when changing the units. I don't know how to reproduce this problem, it seems to work as expected for me.

from the bug reporting page

What not to report here

Ahh, Would it be possible please to make at least VS 2015 supported then? I mean I uninstalled VS 2012 ages ago, Still have VS 2015 as an old code path compiler backup but most VS users will be like me and already moved on to VS 2017. We cant be expected to keep 3 generation old compile paths installed.

We don't currently support looking up these attributes on other objects, only on the current object. Marking this as a To Do. It wold be a good isolated task for a new developer to do if they are interested.