Page 2 of 3
Re: Luxcore vs Cycles Final/Viewport
Posted: Mon Dec 09, 2019 4:00 pm
by Dade
Sharlybg wrote: ↑Mon Dec 09, 2019 3:57 pm
The problem i'm mentioning is very obvious. please just open the file switch to viewport render and just change render engine from luxcore to cycles and back.
See the speed in convergence.
As written above:
posting.php?mode=quote&f=4&p=18452#pr18444
RTPATHCPU is used for view port rendering while PATHCPU is used for final renders. They are 2 different render engines, the first is optimized for fast editing, the second for rendering.
Re: Luxcore vs Cycles Final/Viewport
Posted: Mon Dec 09, 2019 4:23 pm
by Sharlybg
I know that viewport engine is RTPATHCPU. So what i mean is that you can easily see that :
1/___Our viewport engine is much slower than Cycles viewport engine
2/___ running for longer with F12 engine show the same speed issue : 10X time render test.
Lux
Cycles
Re: Luxcore vs Cycles Final/Viewport
Posted: Mon Dec 09, 2019 5:09 pm
by Sharlybg
Another test No caching. just an area light. Max bounce = 4 for both. look how strong lux shadow are while cycles look more softer. Not saying that cycles look better just wondering if this have to do with some kind of trick under the hood ?
Luxcore
Cycles
File
Re: Luxcore vs Cycles Final/Viewport
Posted: Mon Dec 09, 2019 5:54 pm
by Sharlybg
Just doing more test and look like Luxcore convergence can be really affected by poor adaptive settings :
Both are luxcore see how they clearly match cycles convergence quality :
Default
Customized
Cycles
Re: Luxcore vs Cycles Final/Viewport
Posted: Mon Dec 09, 2019 6:04 pm
by Sharlybg
Just applied the same knowledge of adaptive on the porsche scene. small improvement but Cycles is miles ahead. like we miss something in open 3D space.
same time/CPU
Lux
Cycles
Re: Luxcore vs Cycles Final/Viewport
Posted: Fri Dec 20, 2019 7:26 pm
by Sharlybg
https://forum.appleseedhq.net/t/conveyo ... leseed/665
This open space performance issue seem to be real problem. A comparison between applesseed and luxcore show some already seen pattern.
Luxcore 3h vs applesseed 30mn.
Will be nice if Dez help us look into the file.
Re: Luxcore vs Cycles Final/Viewport
Posted: Sat Dec 21, 2019 4:41 pm
by Fox
10 min with different russian roulette settings, top is extreme rr and bottom has rr off.
Re: Luxcore vs Cycles Final/Viewport
Posted: Sat Dec 21, 2019 6:15 pm
by Sharlybg
Thanks for your input Fox. Dade also mention RR above (page 1).
Dade wrote: ↑Mon Dec 09, 2019 3:35 pm
Sharlybg wrote: ↑Mon Dec 09, 2019 3:23 pm
Note that shader are the same.
Does Cycles have Russian Roulette ?
Anyway, an analysis must start from simple cases: use a point light and use direct light only (and don't run the test for only 7 seconds or even a microscopic difference in export/start up time can make a huge difference in only 7 seconds).
Russian roulette isn't exposed in blendluxcore i think. Also if it is the key feature affecting rendertime like that can we make this feature adaptative ?
Re: Luxcore vs Cycles Final/Viewport
Posted: Sat Dec 21, 2019 6:30 pm
by Sharlybg
Re: Luxcore vs Cycles Final/Viewport
Posted: Sat Dec 21, 2019 6:35 pm
by Fox
The rr is little biased, but it's not strongly visible in other scenes than heavy complex light paths. For an example my camera scene with solid glass lenses, the aperture blades were behind 5'th lens and they became invisible with extreme rr.