Page 2 of 4

Re: wrong light groups with OpenCL

Posted: Tue Mar 05, 2019 5:43 pm
by pixel_peeper
update:

I downloaded this buid and the bug seems fixed, thank you!

the only thing that doesn't work properly with OpenCL is this one, some lights are darker compared to other identical. I don't know how to explain in a better way but if you try to render this scene with OpenCL and CPU and check the passes you will notice it.
pixel_peeper wrote: Tue Mar 05, 2019 9:37 am Ok, if I disable clamp everything works fine on the scene I uploaded in this thread. But with my project scene the strange behaviour of some emitter that remains dark or with fewer samples is still present.

Here there is a simplified version of my project scene, the fewer samples issue in some random instance could be seen more clearly in Col-R and Col-B Render Layer.

https://ufile.io/huo7v

If anyone has also any tip on how to reduce render time (and fireflies) is much appreciated :D
again, thanks for the fast support!!

Re: wrong light groups with OpenCL

Posted: Tue Mar 05, 2019 8:42 pm
by Sharlybg
If you follow the workflow describe here you should not have problem with to long render time :

https://www.google.com/url?sa=t&source= ... m8ez2zTVkl

Also not that the more light group you have slower will be your performance.if possible try to use PGI cache.

Re: wrong light groups with OpenCL

Posted: Tue Mar 05, 2019 9:41 pm
by pixel_peeper
I need to reproduce a very realistic array of SMD LED, many many lights that are passing through a translucent material, I know that's a very difficult scene for a ray tracer. But I think that I will change the whole approach using a mix of Cycles, LuxCore and some compositing trick, I Just need to figure out how to achieve the desired result.

Re: wrong light groups with OpenCL

Posted: Tue Mar 05, 2019 11:22 pm
by B.Y.O.B.
If each light only illuminates a small area, DLSC (direct light sampling cache) could also be interesting for you. It is available in the light strategy dropdown.
viewtopic.php?f=5&t=430
viewtopic.php?f=5&t=675

Re: wrong light groups with OpenCL

Posted: Wed Mar 06, 2019 8:57 am
by Sharlybg
pixel_peeper wrote: Tue Mar 05, 2019 9:41 pm I need to reproduce a very realistic array of SMD LED, many many lights that are passing through a translucent material, I know that's a very difficult scene for a ray tracer. But I think that I will change the whole approach using a mix of Cycles, LuxCore and some compositing trick, I Just need to figure out how to achieve the desired result.
Can you post a picture of the kind of image you want to achieve ?

Re: wrong light groups with OpenCL

Posted: Wed Mar 06, 2019 9:35 am
by pixel_peeper
B.Y.O.B. wrote: Tue Mar 05, 2019 11:22 pm If each light only illuminates a small area, DLSC (direct light sampling cache) could also be interesting for you. It is available in the light strategy dropdown.
viewtopic.php?f=5&t=430
viewtopic.php?f=5&t=675
I've tried but I can't get a significant improvement (and I noticed a random color shifting in some lights), I'm sorry that I can't investigate more this function but since I'm a bit hurry I need to finish this render in other ways. Thank you!

PS: just for the benefit of the software and the community please try to verifiy if this is a real bug or just a wrong setting due to my low experience with LuxCore, thank you! Here the post I'm referring to: viewtopic.php?p=10983#p10983

;)

Re: wrong light groups with OpenCL

Posted: Wed Mar 06, 2019 9:42 am
by pixel_peeper
Sharlybg wrote: Wed Mar 06, 2019 8:57 am
pixel_peeper wrote: Tue Mar 05, 2019 9:41 pm I need to reproduce a very realistic array of SMD LED, many many lights that are passing through a translucent material, I know that's a very difficult scene for a ray tracer. But I think that I will change the whole approach using a mix of Cycles, LuxCore and some compositing trick, I Just need to figure out how to achieve the desired result.
Can you post a picture of the kind of image you want to achieve ?
Yep! Like this, but with more detail on the LEDs that change their intensity during the animation, so you can clearly see the diode wich increase its brightness.

Image

Re: wrong light groups with OpenCL

Posted: Wed Mar 06, 2019 4:44 pm
by FarbigeWelt
pixel_peeper wrote: Wed Mar 06, 2019 9:42 am
Yep! Like this, but with more detail on the LEDs that change their intensity during the animation, so you can clearly see the diode wich increase its brightness.
Instead of light groups you can change the intensity of your material emiitting light.
Set a key for intensity and animate the value.

The simplest shape for the LED plane can be a triangle. This reduces number of faces emitting light to one.
If you have circles the number of lights depends on the the points per circumstance.

Re: wrong light groups with OpenCL

Posted: Wed Mar 06, 2019 4:52 pm
by pixel_peeper
FarbigeWelt wrote: Wed Mar 06, 2019 4:44 pm Instead of light groups you can change the intensity of your material emiitting light.
Set a key for intensity and animate the value.

The simplest shape for the LED plane can be a triangle. This reduces number of faces emitting light to one.
If you have circles the number of lights depends on the the points per circumstance.
Unfortunately Blender itself doesn't allow this https://github.com/LuxCoreRender/BlendL ... issues/167

My LEDs needs to be very similar to these ones so I have a quite accurate lowpoly model, now I'm getting the desired result through a mix of Lux, Cycles end some compositing.

Re: wrong light groups with OpenCL

Posted: Wed Mar 06, 2019 5:27 pm
by B.Y.O.B.
pixel_peeper wrote: Tue Mar 05, 2019 5:43 pm the only thing that doesn't work properly with OpenCL is this one, some lights are darker compared to other identical. I don't know how to explain in a better way but if you try to render this scene with OpenCL and CPU and check the passes you will notice it.
I can reproduce this at the 128 samples you set in the file, even with OpenCL CPU device (to eliminate possible driver issues). There might be a bug in the OpenCL code.
Have you tried higher sample numbers (something like 1000)?