9700 and 128 bit color

[snyper1982]

ok i had totaly forgot about this card using 128 bit color. I was wondering how to actualy use it though. does the software have to support it? or is it a hardware feature? im just curious here guys. thanks.

[mamisano]

Ummm...It does not support 128bit color....only 32 like all current cards. The GeForce FX is supposed to support 64 and 128 bit color.

[snyper1982]

take a look and tell me it doesnt support 128.
http://www.tomshardware.com/graphic/20020718/radeon9700-06.html

[mamisano]

Originally posted by snyper1982
take a look and tell me it doesnt support 128.
http://www.tomshardware.com/graphic/20020718/radeon9700-06.html

Well, I stand corrected. I am not sure how that works. I wonder if programs/games need to be programmed to support 64 and 128bit. All I know is the latest drivers do not have support for them.

[TruckChase!]

It's floating point color for internal calculations, the output will still be 32bpp. (same for GFFX) I'm not sure if it's enabled by default, but I doubt it since it would incur a large performance penalty. As for enabling it, I'll figure it out after i get my 9500 and mod it to a 9700. :)

Here's the concept of internal floating point calculations btw: (up to 128bit precision)

Multitexturing is the best example. Say you lay one texture on top of the other, and the top tex is semi-transparent. When you figure the color, you start with 8 bits of precision for each color for each texture. When you overlay the colors, they must be averaged to attain the visible color. Now you've taken each 32 bit texture (effectivly 64 bits of info) and effectively reduced it to one 32 bit texture. Now each time you lay on another texture, you loose that much more info, so the more layers, the more precision you loose.

By doing internal FP calculations, you are able to accurately depict the color combiniations and such, then downsample as necessary. (most likely to 32bpp) The overall color output, even when writing to a 32bpp frame buffer should be far more accurate. These calcs are also applied to non-color operations. (z-buffer checks, etc.)

The kyro 2 chip was similar to this for it's time: It did all internal calculations at 32 bit even when output was set to 16bit. That's why it's 16bit output was far superior to Nvidia's, Ati's, and arguably better than 3dfx's "22-bit" output.