4 Gedanken zu “Kray 3 – VPR Demo

  1. G.Tanski about GPU render in future Kray 3.

    About GPU rendering.

    I don’t like to waste rendering time, so naturally is I want to use as many computing power as possible. Kray will definitly go for the GPU. Actually there is a lot of work done already and Kray is able to raytrace on GPU via OpenCL. Pure raytrace, this means no shading, just computing ray-geometry intersection (and you can test it even inside LW, but some comand line is needed advanced users can type „help gpgpu;“ „help opencl;“ „help raycast;“ in Kray Commander inside LW and play with those commands) . It is too little for rendering anything, but from my point of view lots of problems are solved already. Mechanism for transfering geometry and raycache into GPU works and render hard core works too (ray-geometry intersection).

    GPU support is not priority now for two reasons. One, we have bugs to fix and this is priority now. Two, I would like to implement shading and rendering algorithm in OpenCL 2.1 which allows compute kernels to be written in C++. This speeds up developement a lot and cost of future maintance. I choose OpenCL because its cross vendor standard and when for example you have GPU on motherboard + external graphics card it can utilize both of them and CPU in addition.
    EDIT: OpenCL 2.1 is quite young and not well supported yet.

    In CPU vs GPU discussion I will say I would like to have both supported. There are some tasks that works faster on GPU, but some different tasks are possible only on CPU. For global illumination scene with many bounces (interiors). CPUs flexibility and smarter rendering algorithms (reusing data, using aproximations) gives more benefits then GPUs brute force power. For brute force setups of simple lighting (not many light bounces) GPU will beat CPU.


Schreibe einen Kommentar