[Bf-committers] "Official" CUDA Benchmark/Implementation Thread

[Yves Poissant]

From: "Marko Radojcic" <sambucuself at gmail.com>

> I think that best GPU acceleration performance is achieved if we do
> the work in layers...
>
> First we generate all (or most rays) and then do all the traversing,
> then we do all the intersections and then all the ray collection -
> this is just a rough sketch ...

>From my current experience, you are definitely on the right track with 
approaches along those lines. I would just replace the word "all" by 
"chunks" or "bundles". "All" is going too much toward the other extreme. But 
"bundles" is a good compromise and the bundle size can be adapted to the 
available hardware.

Yves