Light pre-pass: which is faster 2 x RGB 11 11 10 or RGBA 16F?
category: code [glöplog]
On a light pre-pass renderer which is faster, as both are 64bits per pixel:
using a single RGBA16F to store diffuse + specular intensity
or
using 2 RGB 11 11 10 targets, storing diffuse and specular separately
using a single RGBA16F to store diffuse + specular intensity
or
using 2 RGB 11 11 10 targets, storing diffuse and specular separately
Try it out?
I did, but I didn't make much of a conclusion. Having a single machine to test also doesn't help having realistic results.
Make a logging test-app, and we'll run it.
if you made as much productions as you asked questions...
...you wouldn't thumb them up.
Depends on how RBE limited you really are ofcourse, but beware of architectural differences if testing: http://www.hardware.fr/medias/photos_news/00/35/IMG0035597.gif
write once and in effect read once is always faster on the cache level.
