No 32 Bit in 3D Just 2D. Is this true?

[Mike H]

Although it’s true that 32-bit color modes are available for displaying desktop environments (2D), as yet, there are NO graphics cards that can produce 32-bit color in 3D games (and that includes the G400 and TNT2).

The term ‘32-bit’ is slightly misleading when applied to 3D-graphics. What 32-bits actually refers to is the internal data representation that typically consists of 24-bit color (8 bits each for red, green & blue) plus 8-bits for alpha (transparency). In fact most modern 3D cards have this type of ‘internal’ set-up including the original Voodoo chip-set designed in 1995!

When people refer to the ‘color depth’ of an image they are normally talking about the color depth within the frame-buffer (where the final image is stored) and this typically contains only 16 or 24-bits of color information. When companies like NVidia refer to a 32-bit frame-buffer they are actually referring to the extra 8-bits of 'stencil-buffer' that their products (the TNT1 & TNT2) support.

So strictly speaking… this ‘32-bit verses 16-bit’ debate is actually a ‘24-bit verses 16-bit’ debate!

From this webpage

http://www.fullon3d.com/hardware/articles/0899/c_depth/



------------------
Mike H.

Watercooled AMD K6-3 450 @ 500 (100*5) Epox EP-MVP3G Rev. 1.0 Bios 7-16-99 128 Meg PC100 Matrox Marvel G-200 TV 16MB 5.15 Drivers SB Live! Full Retail Western Digital Caviar AC313000RTL

[SteveC]

Yep - AFAIK that's true.

------------------
Cheers,
Steve

My PC? Not that bad, got all sorts of crap in it, and all sorts of crap around it and my desk is also messy. Now what does that say about me? ;¬)

[Maggi]

bsically that is correct ...

but when you use 16bit textures that contain an alpha channel for transparencies, you stumble into trouble.

There are plenty variances for the colordepth in each channel of 16 bit textures, for example RGB: 5-6-5, 5-5-6 and 6-5-5 (bits per channel, meaning either 32 or 64 steps in luminosity)

RGBA: 4-4-4-4 & 5-5-5-1 are the most commonly used format, e.g. you either have 12 bit color information & 4 bit (16 colors) for transparencies or 15 bit colors and only one bit for transparencies, that leads directly into ugly aliased borders at the egde of the tranparency information.

Now think about 24 & 32 bit ...

RGB: 8-8-8 > 256 steps per channel resulting in ~16.9 millions possible colors (vs. 65536 when using 16bit)

RGBA: 8-8-8-8 > 16.9 milions colors plus 256 steps for the transparency information.

...

To me it's pretty clear that 16bit output has NO future at all.

It was an accepatable comprimise a couple of years ago, but since I saw 32bpp on my former G200s, I refuse to accept the term 'visual quality' in use with HiColor !

...

I'll leave the other variances of 32bpp to the in-deep technicians (stencil buffer and stuff) and I hope I could shed some more light on this topic.


------------------
Cheerio,
Maggi

Asus P2B-S @ 112MHz FSB - Bios 1009 final
Celeron300A @ 504Mhz
128MB 7ns SDRAM
G400 DualHead 32MB SGRAM @ 201 MHz memory clock

[JamesA]

24bit colorspace has been labeled truecolor for a very long time. If I am not mistaken, 24bits is "good enough" for the human eye. Anything more is a waste of bits.

Of course, the usage of the colorspace is important. RGB format is one example that has too much allocation (with respect to the human eyes bias sensitivity in favor of green and red) in the blue region.

I think the YUV, and other non-RGB schemes that focus more on hue, chrominance, luminance, et al have a better representation of human color perception.

Took a graduate class in computer graphics, and I learn human biology!

BTW, I am slightly red-green color blind (and, yes, I can read traffic lights fine), and my favorite color is blue (the color human eyes are least sensitive to).

Off to munch, munch with wifey.

JamesA.

[franksch3]

I'm starting to puke from all those hardware review sites. They never have anything completely correct.

First of all: In 2D you don't have 32 bit COLOR. You have a 32 bit framebuffer where for each pixel 24 bit of color information are used. The remaining 8 bits per pixel in the framebuffer are of NO use. They are only there to speed up the drawing process, since memory accesses are MUCH FASTER when 32 bit aligned then 24 bit aligned. So it's a tradeoff between memory usage and speed.

So both in 2D and 3D a color in the framebuffer is represented by 24 bits!!! 8 bits for each color component (red, green, blue).

Secondly: In the 3D framebuffer there is only the color information. So for each pixel we use 24 bit and 8 bit which are not used but only serve to speed up things.

However for textures we can use the extra 8 bit for alpha information, eg. transparency.

The 8 bits not used in the framebuffer are NOT used for the 8 bit stencil buffer. The stencil buffer is combined with the Z-buffer. On all cards with a 32 bit z-buffer we have the following possibilities:
- 32 bits for each Z-value, 0 bits for stencil value
- 24 bits for each Z-value, 8 bits for stencil value

However what he is right about, is that the real debate is about 16 bit COLOR versus 24 bit COLOR.

I hope this explains this enough, because I have seen numerous reviewers make those mistakes and then make the most rediculous statements. If you don't know how it works than only present the bencmark numbers, but don't try to make any statements about it.

With regards
Frank Schoondermark

[franksch3]

This was not meant for the previous two posts since they apeared when I was typing this. ;-)

[Himself]

Yay, another warrior in the fight for common sense.

[franksch3]

I did notice however that the author has changed his 'typo' about the 8 bit stencil buffer in the framebuffer part.

[mragsdale]

up

[Buuri]

One shouldn't be too aghast about this while a normal television can produce only about 2 million different shades of color, so 16 million should be sufficient even for the best of monitors!

B

[Agent31]

yeah but Buuri, that's probably only using the crappy NTSC television standard which so many people have to suffer with, I'm quite sure PAL viewers would see more colours on the screen than that... although I could be wrong, but NTSC still sucks.

True 24/32bit is the future, when memory bandwidth increases in size and decreases in cost, no doubt we will all see and feel the benefits... ooooh, feel that colour!

[Buuri]

I'm Finnish and living in Finland, which can be read as PAL. I know the quality in NTSC is worse.
I don't know the detailed numbers but about 2-3 million shades is close to what an analog TV system can produce at the best.

B

[franksch3]

Somebody didn't read the thread about how to avoid double posts. Oh, wait... This was a tripple post. ;-)

And by the way, a television only shows around 200000 colors.

[Maggi]

Hi franksch3 ... sorry !!!

My browser told me that 'an error occured',
thus I submitted a few too much

...

back on top ... fukking DF31 & loserants

------------------
Cheerio,
Maggi

Asus P2B-S @ 112MHz FSB - Bios 1009 final
Celeron300A @ 504Mhz
128MB 7ns SDRAM
G400 DualHead 32MB SGRAM @ 201 MHz memory clock

[Helmchen2000]

up

------------------
and ty%&§!° went down again

Helmchen