GsoC: Raytrace optimization test

[Ace Dragon]

I can't wait to see how scenes with a lot of reflections/refractions will be sped up from this project, it would be nice to actually do scenes with areas of reflecting/refracting hair particles (provided there's practical use like glass fibers), without needing a mainframe to get reasonable rendertimes.

[Sienio]

Some interesting information about gsoc:
http://lists.blender.org/pipermail/soc-2009-dev/2009-June/000073.html

hmm... I just started thinking if the array modifer could support instancing?
Anyway jagurandi said he needs some scenes to test features(list in link), so will be nice if someone find/do this scenes

[Ace Dragon]

The Blender website and the projects site are both down right now, in fact they were up only at one time all day today

[Sim88]

The mailing lists are readable (and have been so the whole day), it is the link that is not correct.

[Sienio]

Originally Posted by Sim88 The mailing lists are readable (and have been so the whole day), it is the link that is not correct.

ooops... I correct the link, Thanks!

[oogsnoepje]

I've just tested this build with the b.m.p.s. 1.5, but apparently it's too hardcore for this project. Render times with the water material are actually slower

If the developer would like to play / test with it, here's the link:
http://blenderartists.org/forum/showthread.php?t=151903

Select the object and assign, for example, the water material to it.. lots of fun!

[ZanQdo]

I'd like to proclame a new record of number of polygons ever rendered in Blender

1.15 billion tris with blurry reflections:

http://www.zanqdo.com/tmp/Dragon_1.15Billion.png

Rendertime 23 min, Mem 200 mb

[ZanQdo]

Originally Posted by oogsnoepje I've just tested this build with the b.m.p.s. 1.5, but apparently it's too hardcore for this project. Render times with the water material are actually slower If the developer would like to play / test with it, here's the link: http://blenderartists.org/forum/showthread.php?t=151903 Select the object and assign, for example, the water material to it.. lots of fun!

that cube that encloses the whole scene, subdivide it a few times, having such large faces is bad for optimization

[oogsnoepje]

Great idea! *hits himself for not noticing that

Will try again later

[Sienio]

Originally Posted by ZanQdo I'd like to proclame a new record of number of polygons ever rendered in Blender 1.15 billion tris with blurry reflections: http://www.zanqdo.com/tmp/Dragon_1.15Billion.png Rendertime 23 min, Mem 200 mb

zOMG! Working with this polycount must be horrible, but the rendertime and Ram usage isn't bad.
Hope the another gsoc project - openGL viewport optimization(VBO), makes the performance better, there are a few commits of this branch, so if any interesting commits appears I'll build this branch too

Cheers

[Tea_Monster]

Originally Posted by ZanQdo I'd like to proclame a new record of number of polygons ever rendered in Blender 1.15 billion tris with blurry reflections: http://www.zanqdo.com/tmp/Dragon_1.15Billion.png Rendertime 23 min, Mem 200 mb

Holy Crap!

[Ace Dragon]

This could end up being the best thing to ever happen to the renderer (alongside all the work Farsthary and Broken has done of course)

Having all the features devs. like Broken put into the raytracer is one thing but having it render at a fast speed with over a billion tris is another thing, this is excellent work and it has a proven GSoC dev. behind it as well (Jaguarandi already did the successful shrinkwrap project)

Now combine it with the other project slated to greatly increase viewport speed and 2.5 will just drip with awesome, considering it would have a greatly sped up raytracer, viewport, and GE, and will be the perfect foundation for projects like photon mapping and volumetric rendering to formally be included in the official release.

[oogsnoepje]

Originally Posted by ZanQdo that cube that encloses the whole scene, subdivide it a few times, having such large faces is bad for optimization

Okay, tried that, but it still doesn't really seem to render any faster.

[Ace Dragon]

Seemed like Blendernation picked up the test, and a few thought it wasn't even close to impressive because it's just the same object in a grid array.

There should be a test using a more practical scene, how does the raytracing overhaul hold up when using a practical scene one would use in production?

[egan]

Originally Posted by Cyborg Dragon Seemed like Blendernation picked up the test, and a few thought it wasn't even close to impressive because it's just the same object in a grid array. There should be a test using a more practical scene, how does the raytracing overhaul hold up when using a practical scene one would use in production?

The funny thing about their complaining was that ZanQdo's render was not possible without the new renderer features, so whether the semantics of "most faces" or "would have been most faces" is most valid is mostly moot.

Personally, I think these improvements are totally awesome, and hope that they will be merged with trunk eventually unlike lightcut's sad demise.

[ZanQdo]

If someone can't see the value of rendering thousands of fully raytraced instances fast and with low memory usage he has much to learn

[Sim88]

I hope particles work that way too

[ZanQdo]

Particles use dupligroups yes!

[YellowLambo]

What computer did they render that on!?! 23 mins, 200mb of RAM? Say what?!?

[cekuhnen]

maybe you want to make a glass bottle test with refractive values and compare that
with the VRay test render.