You have to disable camera ray visibility if you want it to behave like a light. But that’s not something you want for all shadow catcher objects. Other than that it looks correct to me?
Yes, sorry, the render results is correct. I forgot some stuff to do with colour math and the test in the blend file wasn’t a proper test.
I tested it again today after updating the build. I give up. Now I am more inclined to believe that Cycles X is just not optimized for my lower end GPU (MX130), I hope it will be at some point, but for now I give up.
Here is my test, because of previous feedback about the light bounces not high enough, I changed all in Light Path settings to 128. I render twice for each (Var. 1 and Var. 2), the results for the two times are close, so I am only uploading the faster of the two
Another “bug” I’ve found and was wonder if anyone else was experiencing was this:
When OptiX denoising is enabled in the viewport and the input paths are set to “Color+Albedo+Normal” and you look at the sun in the “Sky texture” node attached to the world shader, it produces a black square.
This also occurs if you have a object that is exceptionally bright (E.G. an emission shader set to strength of 100000). The OptiX denoiser in the viewport also seems to have other artifacts with the input path set to “Color+Albedo+Normal” . This is all in comparison to Blender master. Here’s a few screenshots:
Note: I thought this was a bug with my builds of Blender. But it also affects the build from https://builder.blender.org/download/branches/ so it’s not just my build of the Cycles-X branch. Blender master also does not have any issues. Both the Cycles-X branch and master builds done by myself used OptiX 7.2.
Operating system: Linux-5.10.0-6-amd64-x86_64-with-glibc2.31 64 Bits
Graphics card: NVIDIA GeForce RTX 3070 Driver version 465.27 (Also occurred on 460.XX)
Interesting, reminds me of an “optix denoiser exposure bug” i reported way back, that has been marked as “Known Issue”. Let me find the link
https://developer.blender.org/T77552
I thought it only happens with low exposure value in the color management panel. But in your case it seems to be also the case for bright objects, interesting
2 Likes
Thinking back on it now, it’s possibly related to that. Part of the speedup for Cycles-X comes from changes that that seem to affect the scheduling of work for the GPU. But the MX130 might either be too slow or have too few shader modules to work well with the new scheduling and as a result may run slower.
Disclaimer: This is just speculation on my part. Do NOT take any of it as fact.
1 Like
Please always provide context. What SSS algorithm did you use?
Remember Christensen-Burley will be rendered as random walk in CyclesX.
1 Like
@brecht, have you seen MagicaVoxel? That is a voxel editor and a very fast GPU pathtracer by ephtracy. If you haven’t, maybe you would be interested in looking ‘under the hood’ at the kind of optimization tricks used there. Program and its source code available on github. I did a few tests just for fun. I used ‘Room’ in MagicaVoxel and ‘Manchester’ in Aerialod (interactive path tracing renderer for height maps by ephtracy). In both of those I used custom 2k HDRI map, set bounces to 3, samples to 1024 and resolution to 1920x1080. Then I rendered the same scene with similar settings with Cycles X. Here are the results:
‘Room’
Cycles X: 2:14
MagicaVoxel: 0:40
‘Manchester’
Cycles X: 2:42
MagicaVoxel: 0:48
3 Likes
I believe both issues are resolved by eae783e3ce3.
1 Like
Has anyone seen weird issues with light portals in the latest build? I had to revert back to get a render out, so I don’t have a screenshot right now, but I got big bright squares where the light was hitting other surfaces (with the squares conforming to the angles of the surfaces they were on). I’ll get a screenshot of it later after my renders are done.
That resolved the issue, thank you.
I tried to follow CGMatter’s procedural beachball tutorial last night and found the vector Normalize node, which in 2.92 turns the subdivided cube into a sphere, resulted in a glitchy mess. To be expected in these early stages I guess…
Weirdly if I unplug and re-connect the displacement I get a pair of procedural buttocks!
Changing any settings on the shader results in the glitchy shape again:
I posted the link at BlenderArtists and apparently it worked fine for another user:
https://blenderartists.org/uploads/short-url/wI0N1KangHoXDWc9wWUD60qIhTz.blend
This is on Windows 10 with a Quadro RTX 5000 using cycles-x-blender-3.0.0-c77529091f8d-windows64 from 7th May
@sergey I was looking at another project and came across some issues. I’m not 100% sure of the cause, and I’ve had a hard time re-creating the bug from scratch. From my understanding the bug requires the use of a image texture to cause it, and I’ve tried a handful of images yet none of them seem to produce the bug. The image texture I do have that produces the bug is however not mine and I do not believe I can share it publicly, so I’ll share it with you and @brecht in a private message here on devtalk.
Disclaimer: This may not be a bug at all, but instead just me compiling my build of the Cycles-X branch incorrectly. I just don’t know for sure and hence I am sharing the bug.
System Information:
Operating system: Linux-5.10.0-6-amd64-x86_64-with-glibc2.31 64 Bits
Graphics card: NVIDIA GeForce RTX 3070 Driver: 465.27
Blender version: 3.0.0 Alpha, branch: cycles-x, commit date: 2021-05-10 10:36, hash: rBeae783e3ce36
Description of bug:
The denoising normals pass for OIDN and OptiX seem to have issues with a object with this material:
Here are some examples of the issues I’m seeing:
-
If OptiX GPU rendering in the viewport is used and OptiX denoising is enabled with “Color+Albedo+Normal” as the input pass, then visuals like this occur (NOTE: CUDA and CPU viewport rendering with OptiX denoising is unaffected. Only OptiX viewport rendering and OptiX denoising are affected for some reason):
-
When doing final renders (Press F12) and observing the “Denoising Albedo Pass” generated when the “Final render denoiser” is set to either OptiX or OIDN produces results like the one shown below. In this case, CPU and OptiX rendering are affected, CUDA doesn’t appear to be affected (but I may be wrong). One thing that may or may not be of importance is that the percentage of the image affected by the “bug” changes between rendering backends and denoiser. In some cases the difference is small, in others it’s large. (Note: Pixels that appear white actually have values like R=58069701825691327135744, G=-17846559627935716927210520576, B=-748449044461630787329982464)
I’ve tested the new version (eae783e3ce36) and updated the chart (red - new):
Versions tested:
- E-Cycles (2.92)
- Cycles (2.92)
- Cycles (2.93b) (9cdf11676ecd)
- Cycles (3.0a) (3185084efbe4)
- Cyclex X (1dea1d93d39a)
- Cyclex X (43789b764d9a)
- Cyclex X (a176118b287f)
- Cyclex X (c77529091f8d)
- Cyclex X (eae783e3ce36)
5 Likes
In Cycles X, I wonder if we are going to have a ‘'adaptive world’’ lighting option on world settings (its called adaptive dome lighting or adaptive environment lighting etc in other softwares or renderers) to speed up cycles renderings instead of using old ‘‘portals’’ method. It could also really improve render speed as well. (Sorry if that feature already exist in cycles maybe in other name and if ı am not aware of it)
It is fixed now, but it was not specific to the Cycles X. If an issue happens with the currently upcoming release or master branch the issue is to be reported to the bug tracker.
Thank you. Sorry, I wasn’t able to reproduce the bug in the limited testing with Blender master, hence why I reported it here as a bug for Cycles-X.
The Shadow Catcher “disabled” the denoising.
I had posted a bug report on developer.blender.org and then got sent here.
To avoid posting the report twice, here is the link.
https://developer.blender.org/T88221
Interesting project, but as far as I can tell there is no source code for the renderer on the github page so “looking under the hood” isn’t possible. The source code listed in the releases page is for the webpage you see on github which is weird…