I thought I'd do a few tests to try and find the limits of what TG2 can currently do as far as populations go. I'm pretty sure I could go even further than this, judging by the memory footprint, but here's what I achieved so far:
Here are the scene details:
Earth-sized (default) planet
Extreme (15,000 meters) displacement terrain
Object: Shagbark Hickory, 280,000 tris
16,777,000 trees
4.7 trillion tris total
6min 15sec to populate (not included in render times)
1000+km view distance
Detail 1, AA6, GI 2/2, Very High Pop Detail
Peak memory usage when rendering: 3.1GB
Computer System:
Core 2 Quad Q6600
2.4Ghz x 4 cores
8GB DDR2
Windows 7 RC1 (build 7100)
[attachthumb=1]
Full render - 2hrs 4mins
[attachthumb=2]
Without atmosphere - 48mins 6secs
[attachthumb=3]
Without objects (terrain only) - 5mins
[attachthumb=4]
Without terrain (flat planet, for reference) - 4mins 19secs
[attachthumb=5]
3D preview with instance previews enabled!
Notice how little difference there is between the render with and without terrain displacement. TG2 is quite efficient at rendering large amounts of complex displacements. Clearly atmospherics are far more demanding, and hopefully there is more room to optimize there.
Interestingly I actually rendered in 2.0.1.1 initially - the first TG2 final release available. Render times in that version were 3 hours, 30 minutes! Nearly an hour and a half longer. As many of you know we were able to make a significant optimization after release, and there was much discussion about that around that time. It's interesting to see this particular example of it though. It applies very specifically to the atmosphere as the render time without atmosphere was very similar. The rendered result looked very similar visually, with just some slight changes in grain.
Anyway I thought some of you might get a kick out of how far I was able to take this. Maybe I'll make this into an actual, interesting scene at some point...
- Oshyan