[Polyxo]

Hi Developers,
the other day I saw a first proof of concept of Blenders new Render-Engine.
It of course still looks relatively limited shading wise - one thing however impressed me greatly:
The Engine reacted absolutely instantly on changes to Lighting, Object-Visibility, Count, Position and even Shape!
It didn't quite come across - but it might even do all this in in Path-Tracing-Mode (shown towards the End of the Clip).

I wonder - is such unlimited interactivity maybe an advantage of GPU-based approaches?

Thanks for a short Note!

Holger

[Lars Magnusson]

Hm, the first part where he handles the monkey heads are stated in the beginning of the clip to be CPU rendering, like maxwell Fire.
GPU rendering is when he handles the train.
So it probably has something to do with how for example MaxwellStudio transfers geometry and information into Fire ?

Still, the biggest problem with GPU rendering is what NextLimit and others have explained....GPU coding isn't advanced enough to support all types of shaders. In a couple of years though...wouldn't be surpriced if the GPU's actually reach a state when it can do it all

[Maximus3D]

If you are impresssed by Blenders upcoming renderer then have a look at this.. it can make a grown man cry.

http://research.tri-ace.com/detail_1.html

/ Magnus

[Aniki]

Incredible demo, Magnus! Maybe we will get MW 3 for PS4 ;D

[Polyxo]

If you are impresssed by Blenders upcoming renderer then have a look at this.. it can make a grown man cry.

Magnus,
my point is not at all a Render-Quality-Comparison.* Maxwell in this respect already delivers all I'm after.
What I would like to know is: Why does Maxwell need complete Revoxeling on most Condition-Changes while
this early Blender-Prototype can handle these without having to start over.
It was very useful if Maxwell at some point worked this way too.

Holger

*That stylized Game-Graphics stuff you posted is not at all my cup of tea btw.

[Maximus3D]

It's a pretty sweet techdemo i admit that. But then again it's just a techdemo which demonstrates a engine in developement which means nothing to us now and probably never will.

I'm aware of what your post is about Polyxo and sorry for the temporary derailing of it. Perhaps a adaptive revoxelization would be better where it only updates the changes in a scene instead of the entire scene, this would cut down the long delays for the realtime preview and also when rendering animations with Maxwell. As not all models changes position and properties from one frame to the next it's unnecessary to voxelize them over and over again.

/ Magnus

[Aji Enrico]

I wonder if you could do voxelizations on per object base instead of per scene base, cache those objects separately and at the end gather them all into a single collected voxelized scene.
Then you would only need to revoxelize those changed objects and reuse the other previous cached voxelized objects and quicker create a new scene.
Downside would probably be using more data (ram) and initial voxelization would take a little bit more time.
I haven't touched nor looked how voxelization can be implemented but mesh objects can have this approach, so in other words it is just a guess on my part