Got from one minute 45 seconds to 54 seconds! (on 2.49) on the same pathetic PC! (bad news for newegg!)
It is amazing! This people really are geniuses, I bet they can work for Autodesk or MAya or 3dStudiomax (ohh yes autodesk own them sorry, but they wont own blender! God I hop Im right about that) lightwave or whatever but instead they are making code for an open source project, so a geek like me can have some fun, now seriously I swear that if one day I decide to grow from a hobby to a busines (no matter how small) and I make some money, I will send part of it to the blender foundation.
Based on test.blend: 15.21 to 06.61… Less than half the time!
But i had one image that took over an our (lots of fur and a little transparencies) to render with 2.49… and about 26 SECONDS with 2.5.
Obviously some calculations have been massively enhanced!
I rendered LotR Junkie’s stargate scene in 2.49 and at 1000x1000 with OSA 16, it took 1:09min to render. I then opened the scene in 2.5 and rendered it with the same settings and it took 2:07 min. :S
The glass seems to look better in the 2.5 pic.:eyebrowlift:
Obviously some calculations have been massively enhanced!
The biggest speedups should come in scenes that break octrees and make them fail (like the Teapot in Stadium scenario). I had one scene with stone Dragons that rendered quite fast in 2.5 and 6 hours to render in 2.49a, the reason being the layout caused the 2.49a octree to fail spectaculary.
The file is available on these very forums. I don’t know where to post it however, and well it isn’t even my scene so I’m not sure I should be the one reporting the issue?
In 2.5 you can always choose “octree” method of optimization for rendering (In the render settings, look for the “performance” tab, in there the option is “Acceleration method”) so you can get the old speed in your notebook for this model. For what i know, not every optimization engine works with every geometry. The only optimization that works with any geometry given, is to convert the slower parts of the render engine from C/C++ code to assembly. Of course this is not feasible for software like Blender, since must be portable across different platforms…
The more CPUs you have for render the better… Wonder if Blender 2.6 will break the 8 threads limitation. 2010 is the starting year of the 12 cores in a box.:eyebrowlift: