Film - the bottleneck

[Lord Crc]

Builds won't be that useful I think, most who have the serious hardware seem to run some *nix variant for some reason I think I'll set up a couple of branches.

Jens just tested my current patch on his 6 core xeon with hyperthreading enabled, thus 12 threads, on the simple scene you see in the initial post.

hybrid-bidir reference: 1min, 12T, 46.58s/p, 359kS/s, 266%, 105MC/s, GPU Load 7%
hybrid-bidir my patch: 1min, 12T, 77.08s/p, 655.03kS/s, 266%, 174MC/s, GPU Load 13%
plain bidir reference: 1min, 12T, 63.50s/p, 548.63kS/s, 294%, 161MC/s no-hyb
plain bidir my patch: 1min, 12T, 76.55s/p, 661.35kS/s, 294%, 194MC/s no-hyb

So roughly 65% increase in the hybrid case and a fair 20% increase for plain sampler renderer.

[SATtva]

Christ! Yeah, fewer the cores, less the benefit, but still this is a huge improvement, especially with GPU-enabled rendering. Great work Lord Crc!

[Lord Crc]

Thanks Yeah quite noticeable improvement. Would be fun to see how it performs on bigger iron.

Pushed the changes into a new branch, see http://src.luxrender.net/lux/rev/431ddf172614

[B.Y.O.B.]

So roughly 65% increase in the hybrid case and a fair 20% increase for plain sampler renderer.

Wow, thank you very much for working on this
Hopefully this will help hybrid bidir to become more valuable for "production use".

[jeanphi]

Hi,

Not bad at all With such numbers, you could have pushed your changes to the main branch at once!

Jeanphi

[Pilchard123]

How does this scale with scene complexity? I hate to put a damper on this, but it may have been such a large improvement because there are large areas of the image that have very similar lighting. Of course, I'm just thinking out loud here, not knowing how the engine works at all - it may be that it gets even better with more complex scenes.

I've asked this before, but what language is Lux written in, and how can I help? I can program passably, and if all else fails, I can just go through the code and add some comments. I remember hearing that there was a lack of those. Finally, I'd suggest making a how-to-help section, with a n00b-walkthrough for setting up a dev environment. Or is there one of those already?

EDIT: C/C++, I should have guessed. Hokay... Still need help with dev environment. Visual Studio okay?

EDIT2: I'll have crack at the GUI "merge FLM" thing if you want. Don't expect miracles, I haven't used any C-variant for a while.

[Lord Crc]

How does this scale with scene complexity?

The improvement will be less if the scene is sufficiently complex. By complex I mean triangle count, complex material setups (many mix layers fex), using procedural textures and similar. Then more time is spent on intersection testing and/or materials, and less in the bottleneck.

I'll have crack at the GUI "merge FLM" thing if you want. Don't expect miracles, I haven't used any C-variant for a while.

Please feel free. If you come on IRC I may be able to help you with questions

[Pilchard123]

Oh, okay. Still, any performance boost is worth having.

[Lord Crc]

Keep in mind that when using hybrid, geometric complexity should be less of an issue, so one should be able to get good scaling with hybrid still.

[Pilchard123]

Sorry to continue the OT, but I was downloading the dependencies with getdeps, and my net connection died. Not I only have half of qt-everywhere-opensource-src-4.6.2 and it's skipped the rest of the download. Any way of getting it?