Pixar tech demo at Nvidia

he covers gpu real time fur,
opensubdiv,
presto
and many other things

[video]http://www.ustream.tv/recorded/45386636[/video]

cheers :slight_smile:

I now have a slight case of computer envy. :slight_smile:

looks like presto also has Z-up :stuck_out_tongue_winking_eye: amazing.

whats most inspiring (hopefully to blender developers too) is that pixar is putting a lot of effort in implementing real time rendering technologies in their pipeline and have good playback speed. Their animators/other artists can see what they are doing, especially on furry characters whose silhouette is dramatically different with the fur on.

Real time global illumination, physically based lighting, dynamic tessellation (based off viewport distance), real time hair rendering (no caching) and so on.

Their final renderer looks like cycles, but way faster

for me Cycles is just so impressive, think that we have brecht and everyone working on this render engine that can compete with katana.
it’s not like BI hopelessly behind but actually on the same path.

first feature complete, then start with the optimization tricks. then real-time. I can live with a OpenGL viewport as long as the final renders are just as awesome as they are with Cycles.

OpenSubDiv is so impressive, I think Dirk even been on this forum, tinkering over the holidays this winter with blender source code regarding opensubdiv.

Their final renderer looks like cycles, but way faster

Are you sure that’s not because of really high end hardware?

oh look its dirk from opensubdiv

well nvidia is there they were like here use our 20 50,000 dollar iray machines haha

Yeah Pixar can afford the best GPUs and software engineers, no surprise there. If it wasn’t Nvidia it would be AMD.

Thanks guys, glad you liked the presentation. The Presto demo was running on a 16 core Sandy Bridge machine with a K6000 GPU (12 gigs, 2800 cores). Most animators here don’t have a GPU that big, we have a lot of Quadro 4000 Fermi (2 gigs, 256 core) GPUs around and a few Kepler based ones. The fur on Sullivan was optimized for Fermi for Monsters University, so we couldn’t afford things like self shadowing etc.

I hope to get back to the OpenSubdiv integration into Blender in time for SIGGRAPH this summer, depending on production pressures here. Really Sergey has done most of the work already.

Dirk, is that 8 core 16 thread or 16 core 32 thread? Very impressive presentation. Presto seems like such a great piece of software to work with.

We don’t see speed gains through hyperthreading in our typical usage, so Presto keeps threads at 1 per core which gives good results for us. In this case then it’s 16 threads on 16 cores. I haven’t checked but we may not use all cores to avoid one taking extra long by sharing with e.g. kernel threads etc- so it could be 12 threads on 16 cores used by Presto.

We have a mix of 12 and 16 core machines currently.

. . . and thanks!

I’m genuinely surprised that all of your animators aren’t rocking top-end Quadros. I would have expected Nvidia to practically force the cards onto you to encourage you to speak out about using their products.