Originally posted by ChrisRay
That is completely dependent on the program your using. And what you are trying to render. There are cases where the extra shading power is relevent,
Whether your creating a nifty screensaver, or Developing a Texture embossing method for a Playstation emulator.
Your mind is narrowed down only to the games we have right now. And as such, your only thinking about games we have today
Uhh from a programmers point of view, 16 bit Floating Point Calculations would be useful. As I said. Its relevent to the subjective situation, In PSX emulation It would preferrable to use texture embossing in a 16 bit fashion. As there would be no benefit to using 24 bit,
In this case, The 300 cannot benefit from the extra speed given by the lower precision.
We're talking about subjective values. To the specific coder.
Never said he ran into limits with its precision, As a matter of fact, I said he preferred 16 bit precision for rendering in Doom 3, Savvy? The interview is there, It specifically says he would prefer 16 bit precision.
And the only reason he did somethings with 24 bit textures is because thats the way the r300 is programmed.
Thats all fine and dandy for you, But alot of people who buy a Geforce 4 Ti 4400 or a Radeon 9500 Pro will not upgrade these cards for like 3-4 years, So yes its rellevent.
Future proofing does exist, In a limited fashion, And most non hardware enthusiasts buy computers because they believe they are future proof.
Using Pixel Shader 2.0 in 16 bit precision to emulate Pixel Shader 1.4 would not be the cause of its current slow down. And theres nothing wrong with emulating 1.4 in 16 bit precision. As theres no benefits for using the higher precision on Pixel Shader 1.4
Ok, so other than a screensaver, or an emulator of a 7 year old console what uses have you got? And you certainly DID allude to carmack saying he ran into limitations using the 24-bit floating point of the R300. Do you even read what you type? You go on with a paragraph about the fp capabilities of the cards, and then say you can see where carmack ran into limitations. Without making mention of Carmack specifically saying he ran into limitations of the NUMBER OF INSTRUCTIONS. I'm also pretty sure R200 supported 8-bit integer max. I also beg to differ that most non enthusiests buy computers because they think they're futureproof...they buy them because they're cheap usually regardless of what's inside. I can tell you've gotta be an 18 year old kid or younger, because you don't know much about the real world, and your ideas go against what has been shown generation after generation for YEARS in the computer world.