Title: Graphics card / CPU collaboration for fast high precision raytracing? Post by: Mrz00m on October 13, 2014, 09:51:40 PM If a graphics card was used to figure ray lengths of a fractal formula to 99% percent precision, and a CPU handled the last 0.01% precision, would the program be able to draw far zoomed in 3d fracals, and would it be 10 times faster, comparable to Kalles Fraktaler program?
Title: Re: Graphics card / CPU collaboration for fast high precision raytracing? Post by: laser blaster on October 13, 2014, 11:17:02 PM Once you're zoomed in close enough to the fractal, single or even double precision will be totally unable to accurately compute even a single ray step. So your idea wouldn't really work out. On the other hand, it might be possible to extend perturbation theory to 3D fractals, allowing for super-deep zooms, but there's no guarantee. For now, I think at least double precision zooms could be practical for a very highly optimized GPU raytracer. No current GPU-based fractal raytracer is anywhere near near being fully optimized. One huge optimization would be to raytrace the scene at a lower resolution, with a higher epsilon value, and use the final depth value as the starting depth for a higher-resolution pass. You could even go up to three or four passes. This would give a large savings in the number of ray steps needed for the final pass. |