R300 & 128bit?

[isochar]

DX 9 requires floating point pixels. R300 is DX 9 compatible. Therefore, R300 will have floating point pixels. A floating point number starts at 32 bits (IEEE single precision float) and goes up from there. 32 bits x 4 (R,G,B, and alpha) is 128. This is where he is getting it from. On the actual display, I believe it is rounded to a 10-bit integer (40 bit color).

The difficulty in implementing floating point pixel paths is why Matrox and 3Dlabs did not put them in the Parhelia or the P10. Both R300 and the NV30 will have them.

The only thing that makes it difficult is the die space required to have 128-bit values and route them around the die. Both NV30 and R300 are huge - it's pretty obvious that that is exactly what they have done, unless they also included huge caches on chip.

Any accuracy behind this?

[Marshmallowman]

I really can't see the point, it must be a convience thing.

but given a floating point value and fix point value of the same bit precision, the floating point maths will incurr nasty errors after doing several calulations (multiplies divides ..etc), whereas fixed point will give produce lower and more consisten error levels.

I can't see the point in using a lot of the floating point stuff used in direct x. Floating point colours just sounds like a load of wank, I know it will probably work its way into more standrds, but there are a lot of reasons it should not.

[Pace]

Floating point colours just sounds like a load of wank

Quote of the day, hehe

[isochar]

Is there a list of DirectX 9 specifications? If so, does it include 128-bit color? (I did a google search but came up empty-handed)