More than 32 bit will be 100% useless.
16 bit color is OK in 90% case.
.. 40bits will be all we need, for a nonlinear colourspace.
then, we will have more than the eye can see, with 32bits (actually 24bits), we're not sure about this. its at the edge.. (40bits == 30bits colour.. or the matrox way, 32bits with 30bits colour).
now, one thing is, we should drop the non-linear colourspace alltogether, it makes much problems and conversions needed. instead, put the gamma correction directly into the screen, as every screen "knows" about its own gamma.
sgi systems work that way.
then, 64bit colour would be enough.
BUT this is just for the ending screenbuffer! don't forget that, to do any pixelshading, we need the same buffers for multipass and for postimageprocessing effects, where we reuse its data. and THAT is where we need much higher precicion, as today, you don't store simply the colour in the colourbuffer, you store positions, you store distances, you store lighting, all you want. its just a buffer. and there we need a simple 32bit float rgba buffer == 128bit buffer.
oh, and, i don't listen to swiss country sound. i listen to trance (i'm even getting a dj
😀).
currently, i'm listening to "dj sequenza - ich schmelze/i'm gone"
and, i'm listening to dido - white flag, too.. beautiful song..
"take a look around" - limp bizkit
www.google.com