Radiosity simulations

[Daishichimaru]

Just curious if anybody actually has the horsepower to use any of the new radiosity simulatons or if I'm doing something very wrong with it. I did some playing with it last night and it makes everything look peachy wonderful, but to get decent results it took my poor machine two hours to compute the solution for a single frame of a 50k poly model with one skylight. Are we doomed to using raytraced omnis and hoping it turns out right or is there some sort of shortcut I'm missing that doesn't involve net rendering?

[GavinO]

I get the solution for a frame of 150K+ polys in 7min for 50% accuracy, 20min for 85%. The 50% looks just fine; don't see what the extra 15min of render time is getting me. I have problems with some of the textures getting way over-saturated, tho. I've got the carpet set to almost black and it comes out as though it were are max. I'm going to have to net render the final, tho, as it just isn't practical to wait overnight for results of a piece of the animation. Currently 12 machines able to be taken for the animation after-hours, amybe more.

[Lev]

If it takes that long to render a single frame with a skylight (i assume you are using light tracer for that), try using radiosity instead. It does take longer to calculate a solution, but once you have it, it can be used for all frames, without re-calculating (unless you have animated lights and such). Combined with area shadows, this can give results very close, or even better than with light tracer+skylight, with significantly lower render times.

Another way is decreasing the polycount. If you have more than 50k faces, you are most likely not using rational and optimised models. A few tips on improving that:
1) Use Optimise and Multires modifiers
2) Use simplier geometry with detailed textures and bump maps. A primitive with a right texture would look better than insanely complex geometry with no maps at all.
3) Most likely not all of the geometry is shown in EACH frame - try temporarily hiding the objects that are not shown (This will especially help alot when using global illumination)

As far as i understand, in most cases, the model with the highest detail, (and polycount) is the robot itself. Try removing any unnecessary detail unless you plan on having close-ups on it. You might actually consider having a separate low-detail model of robot for general scenes, and a detailed one for close-ups.