OctaneRender® 1.024 beta 2.47 (lin/mac/win) [OBSOLETE]

[abstrax]

Thanks for your hard work Refractive. Wonderful release!

I've notice that when i use pmc i can use my pc normally, it's not slow, instead when i use pathracing i can't use my pc for other works

The time of render with pmc is bigger but the possibility to use the pc for others works is great!

Yes the observation is correct that your UI response is more fluent when you use PMC. The reason is that the work chunks are much smaller than in path tracing / direct lighting. We will improve that for those two kernels in the future, too.

Cheers,
Marcus

[abstrax]

Hey very nice, now please, turn caustics off from pathtracing kernel or at least put a switch for this..
It will be great if someone can explain the new sliders in PMC rollout .
I'll try it with some heavy interior scenes very soon - lots of lights and caustics. Also I'll try with exterior scenes - pool scene with lots of lights passing trough water curtains , etc..
Nice job so far.
I notice too the nice speed gain when Octane renders with PMC kernel witch is great.

It's extremely hard to "turn caustics off" as they are basically just indirect light. And you want indirect light in a global illumination rendering, don't you?

To the sliders: There are basically two new sliders in the kernel: "exploration strength" and "direct light importance".

The exploration strength specifies how long the kernel investigates good paths before it tries to find a new path. If you lower it, your image becomes more noisy. If you increase it, it becomes more "splotchy".

The direct light importance makes the kernel focus more on paths with indirect light. For example, imagine sunlight through a window creates a bright spot on the floor. If the direct light importance is 1, the kernel would sample this area a lot, although it becomes clean very quickly. If you reduce the direct light importance the kernel reduces its efforts to sample that area and focuses more on more tricky areas that are harded to render.

Cheers,
Marcus

[abstrax]

thanks, currently testing with an outside night scene full of fireflies. Althou I might be doing something wrong cause I added the disperssion value (0.012 for the glass and 0.02 for water, the glass one I got from wiki, the water is just random) and it's converging faster but it's giving some strange colours on the glass and water, dunno quite why. Also this scene was done with 2.45 I think and light emmiter values are all wrong, so I might have to tweak it further, but I was anxious to see it, lol. I'll post some test later on.
again, thanks.

If you use dispersion > 0 you get the colourful noise. We have some ideas how we might mitigate this effect, but for now you have to live with it. The "colourfulness" vanishes after a while though.

Regarding the emitters: Remember that we changed the power of emitters from power density to absolute power in beta 2.46.

Cheers,
Marcus

[kubo]

thanks Abstrax, yep the color thing it does "clean" somehow after a while, althou I have to say that I like it (yep it looks weird at first but it adds some reality, at least to "my eyes")
I'm trying also an interior scene, and it looks awesome, but I'm going to let them "toast" overnight, so tomorrow I'll post some test just for the fun of it.
No need for the zombies.... but I might try something with them too... arfff you sure know how to bring the fun back to rendering.
Also, in windows 7 x64 2.46b is working fine under cuda 4.0
Thanks, thanks thanks....

[abstrax]

Hey,

first of all big thanks for that pre-release.

I've also tested the new kernel, but not on an firefly scene with much light. The lighting from my testscene is only coming from an hdri.

In my scene, the new kernel is much slower compared to the pathtracing. With PCM i'm getting around 0.54 Ms/sec with my GTX 260, where pathtracing is at 2.00 Ms/sec.

Here the rendertime for 250 samples:
Pathtracing: 00:01:45
PMC: 00:07:46
Directlight: 00:00:42

But as others already encountered the overall system responsiveness is much greater with the new kernel. If the other kernels would achive the same responsiveness is would be awesome. It makes much more fun that way. But maybe this is because of the new cuda version (i also got a second video card installed gtx 430, but i disabled it for octane).

Regarding the responsiveness: We plan to do something about it in the other kernels with the framework rewrite. It may slightly impact the performance though.

Regarding the speed: It really depends on the scene, where PMC is useful and where not. In simple scenes, path tracing has a clear advantage. That's the reason why we keep the other 2 kernels, to let you choose. After a while you will learn when to use which kernel.

Cheers,
Marcus

[abstrax]

Nice job! Octane is able to deliver production render on interior scenes now. There was a indirect light, specular, glossy... on the scene I tested and it was great. Take some time to converge but it does! No fireflies. Thank you! I´m using 2x GTX 470 and it worked pretty fine on multi gpu mode. It is not linear but the gain is very good.
--> don´t know how to use those new controlers exploration_strength, direct light importance and maxrejects
--> I´m using CUDA 4.0
--> The render is a bit darker on pmc mode

Thanks for the image. When you compare the renderings, please rerender in path tracing, too. I forgot to mention that we also fixed a bug in the bump mapping code, which makes things appear darker. If you compare the two images in http://www.refractivesoftware.com/forum ... f=9&t=6735 you can see this effect on the sofas and on the stone of the staircase.

If you still find significant difference between path tracing and PMC, please let us know.

Cheers,
Marcus

[andrian]

It's extremely hard to "turn caustics off" as they are basically just indirect light. And you want indirect light in a global illumination rendering, don't you?

To the sliders: There are basically two new sliders in the kernel: "exploration strength" and "direct light importance".

The exploration strength specifies how long the kernel investigates good paths before it tries to find a new path. If you lower it, your image becomes more noisy. If you increase it, it becomes more "splotchy".

The direct light importance makes the kernel focus more on paths with indirect light. For example, imagine sunlight through a window creates a bright spot on the floor. If the direct light importance is 1, the kernel would sample this area a lot, although it becomes clean very quickly. If you reduce the direct light importance the kernel reduces its efforts to sample that area.

Cheers,
Marcus

Clamp them as maxwell and fry did...
Thanks for the info about sliders..
Anyway here is my tests. Engine "pumps" pixels quite good as I expected

- I notice the bump fix, but now it looks too much (wood and kitchen doors) , compared with older version.

- sun shines trough windows glass, put some nice caustics on ceiling and illuminate the first room, then the light travels trough glossy glass on the doors and trough colored glass as well, which have dispersion and all this, for one hour rendering on GTX 480. Well heavy scenario , impossible for pathtracing engine.

- nice caustics. First appearance was around 14 sec. This is 15 min render.

All in all impressive kernel.

[Jaberwocky]

bit late with these tests anyways , here goes

I did a scene with path Tracing and PMC . Saved both at 1000 and 20000 samples.

See attached the results.I was going for 32000 but it's getting late.so stopped at 20000

timings are also attached for anyone interested

640 x480 pixels
Both pathtracing and PMC were left at default settings

[Jaberwocky]

Sorry, forgot the PMC at 1000 samples

see attached

The result does seem the pmc clears up better , although as you can see from the timings it does currently take longer.

Results rendered on my Gigabyte GTX460 1GB at 715Mhz

Hope this helps Abstrax

[abstrax]

I forgot to exlpain the maxrejects option (don't have the UI in front of me): This slider should usually not be touched. By lowering it, bias is introduced and the rendering is sped up. Should we remove it or leave it in?

Cheers,
Marcus