Scali wrote on 2023-03-06, 11:17:

That is generally in the context of 3D acceleration, where you'd either have 16-bit or 32-bit modes (I don't know of any accelerator that uses the quirky 24-bit mode).
By claiming "internal 32-bit color processing" they mean that any kind of lighting or translucency operations are done with 32-bit precision, even if the final result is reduced to 16-bit before displaying on screen.

I suppose I worded it wrong. I didn't mean precision. I mean are there still 24-bits of RGB color (~16 million) involved during processing, with the extra 8-bits being discarded/unused so to speak, or is the internal color actually 30-bits of RGB (~1 billion) with 2-bits discarded/unused, which then is reduced to 24 or 16-bit depending on the selected output?

eddman wrote on 2023-03-06, 11:51:

I suppose I worded it wrong. I didn't mean precision. I mean are there still 24-bits of color (~16 million) involved during processing, with the extra 8-bits being discarded/unused so to speak, or is the internal color actually 30-bits (~1 billion) with 2-bits discarded/unused, which then is reduced to 24 or 16-bit depending on the selected output?

32-bit in the context of 3D acceleration is generally an 8-bit-per-component format (ARGB).
So you have 24 bits of colour information, and 8 bits for alphachannel, if this is used during rendering.
So when they speak of "32-bit internal" they generally mean that they process everything with 8 bits per component, and then reduce to the final output, which is usually 16-bit.

Scali wrote on 2023-03-06, 11:59:

So you have 24 bits of colour information

Ok, thanks. So the RGB part was always 24-bits max.

I looked into color depth a bit more and apparently nowadays consumer cards can do 30-bit RGB, which I suppose is used only for HDR and whatnot. I haven't yet pinpointed which cards first supported it.

if i'm not mistaken a Matrox Parhelia allowed for 10 bits / color

eddman wrote on 2023-03-06, 16:16:

Ok, thanks. So the RGB part was always 24-bits max.

I looked into color depth a bit more and apparently nowadays consumer cards can do 30-bit RGB, which I suppose is used only for HDR and whatnot. I haven't yet pinpointed which cards first supported it.

Yup, and since Shader Model 2.0, pixel processing can be done with floating point. Most modern GPUs support 32-bit and 64-bit per component floating point.
So internal precision is now usually 128 or even 256-bit. Usually it is then downsampled to 32-bit integer RGB, which is usually 8 bit per component ARGB, but 10-bit is also possible with 30-bit RGB. But that usually has some limitations.

weedeewee wrote on 2023-03-06, 16:24:

if i'm not mistaken a Matrox Parhelia allowed for 10 bits / color

I recall having some kind of Matrox with a 10-bit video option at some point. This would have been Windows 98, I think. There may have been other oddball video modes as well. This was visible from the display properties color combo box.