OTOY Forums

2.01 marginally faster than 2.00, but 1.54 faster still and resolves better.

octane render 2.01 40.62ms noticeably more fireflies octane render 2.00 38.58ms still unresolved fireflies octane render 1.54 45.27ms and much cleaner

MrFurious wrote:2.01 marginally faster than 2.00, but 1.54 faster still and resolves better.

octane render 2.01 40.62ms noticeably more fireflies

2.0.1.jpg

octane render 2.00 38.58ms still unresolved fireflies

2.0.0.jpg

octane render 1.54 45.27ms and much cleaner

1.5.4.jpg

I thought you were talking of a speed drop from 2.00 to 2.01.

Yes, Octane got slower from 1.5 to 2.0. We added a whole lot of new render features which all affected render speed, even if the features are not used. It has been discussed a lot on the forums the last weeks. We are still working through a list of ideas, but there is only so much we can do here, so I can't make any promises that we can improve performance further in the near term.

The trench scene is affected more than most scenes, which is because it's dominated by ray-tracing (it has only one trivial material). Here on my 690 I see even a drop of 18% in path tracing, but in most scenes "only" 5-10%, in some cases 2.01 has the same speed as 1.5.

I'm worried about the noise though: Doesn't look right and we will have a look at it.

EDIT: Some parts are also rendered differently. Only direct lighting seems to be affected.

MrFurious wrote:2.01 marginally faster than 2.00, but 1.54 faster still and resolves better.

octane render 2.01 40.62ms noticeably more fireflies

2.0.1.jpg

octane render 2.00 38.58ms still unresolved fireflies

2.0.0.jpg

octane render 1.54 45.27ms and much cleaner

1.5.4.jpg

Thanks for pointing out the noise. There is a bug in direct lighting that creates that huge amount of noise. It will be fixed with the next release.

Thanks Abstrax. Considering all the new goodies 2.0 brings us a 'minor' performance hit is excusable. Will be glad to see less noise though especially since I almost always use the Direct Lighting Kernel.

Cheers,
Dino.

Only noticing this now playing around with the LW plugin.. indeed tons of noise with 'Diffuse': very simple scene, a single arch model and a single light using DL diffuse. Background noise taking ages to resolve. May need to switch back to 1.54 until this is addressed.

Good to know - thanks
about fireflies, it remains the problem, specialy visbible on glossy/specular surfaces with roughness values close to 0 ( but not =0, where it disappear ). It is not specific to octane but for all GPU renderes ( cycles, thea, which seems to be better for this but not perfect, as far as i've tested ).
It is a very ennoying problem - specialy with animation of glossy surfaces, because render time have to be huge to 'smooth' them ( and the integrated highlight-filters are not the ideal solotion, as it changes a lot the final look of the rendered picture ) and this make loose one of the biggest advantage of GPU renderer: speed.
I'm surprised that no real solution has been found for now around this, it would be a major step...

vipvip wrote:Good to know - thanks
about fireflies, it remains the problem, specialy visbible on glossy/specular surfaces with roughness values close to 0 ( but not =0, where it disappear ). It is not specific to octane but for all GPU renderes ( cycles, thea, which seems to be better for this but not perfect, as far as i've tested ).
It is a very ennoying problem - specialy with animation of glossy surfaces, because render time have to be huge to 'smooth' them ( and the integrated highlight-filters are not the ideal solotion, as it changes a lot the final look of the rendered picture ) and this make loose one of the biggest advantage of GPU renderer: speed.
I'm surprised that no real solution has been found for now around this, it would be a major step...

Try increasing caustic blur.

thanks for the tip, Abstrax, i'll try ASAP

Thanks for the daily build Juanjo.

I look forward to instance surface picking

The latest daily build does not run on Win XP64.

.......entry point GetTickCount64 could not ......

2.01 standalone runs fine however.