Reading Nvidia documentation is says you should use the smallest possible data type to represent vertex attributes so that is what I did… sadly I have found using BYTE(signed or unsigned) or UNSIGNED SHORT seems to run at half speed.
I am working on terrain rendering and when I use signed shorts for the X/Z position of the data I get about 2x the performance of using unsigned shorts, this is with glVertexAttribPointer calls as glVertexPointer only uses signed.
Now I would greatly prefer to use unsigned data as all my values are positive… does anyone know why this speed difference exists? I wasted days trying to figure out why my app was so slow only to find it was just because I was using bytes and unsigned values…
I am using a geforce 6800 btw.
It is possible that the hw does not support the unsigned shorts or does not support the number of components you are using (e.g. if you are using 2 ubytes).
nVidia accelerates only normalized(!!) signed and unsigned bytes and only non-normalized signed shorts.I haven’t been tested integers yet, so my info is only for bytes and shorts.