10 Bit Color ??

[X-Caliber]

Hi,

I read that the new Matrox Parahelian is a true 10 BIT card, but I don't exactly understand what that means. So I guess my G400 is an 8 bit card, but then what does it mean to view a picture in say 8 bit color, 16 bit color, or 24 and 36 bit color? An explanation of all the differences would be great!

Thanks,

-V-

[Kruzin]

Almost
G400 can do 8/16/24/32
Parhelia can do 8/16/24/32/40(10bit gigacolor).
It's only really mildly advantagious in photo editing/photorealistic rendering apps, and most ppl can't tell the diff...but it's there....not a real big selling point...

[DGhost]

can the Parhelia acctually do 40bit color, not just 32bit w/ 10bit RGB, 2 bit Alpha? all the information that I have seen says this is not achievable with the Parhelia...

if it can... wow, i would be very impressed...

[luni]

No, as you say, Parhelia achieves 10 bit per color channel by cutting info from alpha channel. Real 40 bit rendering is possible in Radeon 9700 and forthcoming DX9 cards afaik.

[Novdid]

Yes that is correct DGhost, 10bit RGB and 2bit Alpha.

[az]

Actually, for normal windows use, 24 bit is really the same as 32 bit, it just lacks the 8 bit alpha. So there is no difference if you view a picture in 24 or 32 bits per pixel mode. But (don't know if this is still the case, but it certainly was so a while ago) most cards are actually faster in 32 bit than in 24 bit mode, or there was no performance difference. Had something to do with the way the cards fetched info from their memory IIRC.

AZ

[X-Caliber]

sorry for my ignorance, but what do you guys mean by "alpha"

[thop]

alpha channel is kinda like transparency. defines how much a pixel is visible. 8bit makes 2^8 (=256) states of visibility possible. 0 = invisble, 255 = fully visible. with only 2 bit (2^2) that's only 4.

if i'm not mistaken that is

[DGhost]

Originally posted by luni
No, as you say, Parhelia achieves 10 bit per color channel by cutting info from alpha channel. Real 40 bit rendering is possible in Radeon 9700 and forthcoming DX9 cards afaik.

iirc, the 9700 only has a 32bit framebuffer as well, so it has the same limitation as the parhelia.

[d_constructed]

DGhost, both the NV30 and 9700 can do 128 bit color depth in DirectX land, sure the framebuffer is only 32 bit.
But if some clever dude makes for example a DirectX plugin for Photoshop (or standalone app) that works in DirectX space (that already have 128 or 256 MB FAST gfx mem) and render the output to a file format at 128bit we would see off-the shelf gfc cards outperforming custom high-end apps and hardware.
Maybe someone can forward this idea to a programmer???

[Sinistral]

- I thought it was 128-Bit Internal Colour Presicion. Meaning, Games that optimize the code offer Whiter Whites and Blacker Blacks or Truer Colours. So you get less or no colour banding. Something that isnt a problem in Pro Apps as its all pre-press and not on the fly like games i think.

The RAMDAC's themselves are limited to the 32-Bit Colour Modes. So any output is limited untill new or more advanced RAMDAC's are utilized. -

[DGhost]

iirc, the 9700 doesn't even maintain the 128bit internal precision across the entire 3d pipeline. as with most other graphics cards, the end result is dithered down when written to the frame buffer. it isn't done in DirectX land... its done fully internally to the graphics card chip... you could not simply write a photoshop plugin to work in this color space this level of precision cannot be shown.

if you wanted to keep that level of precision for the output (for final rendering, etc) iit would take a custom bios with custom software (which both ATI and Nvidia are probably working on) that allows you to to write that information... and even then it would be expensive cards and expensive software...

[Pace]

More: fit bit fits fit bit? (for those who speak Aberdonian Scottish ).

Say you have a 1-bit image. That means you can store a 1 or a 0 for each dot on the screen. Therefore, you have a choice of 2 colours - usually a black and white image. If you stretch that to 2-bit, you can have 00, 01, 10 or 11 - so 4 colours. Adding 1-bit always doubles the number of options available.

So, an 8bit image is 256 colours, 12bit is 4096 (remember the Amiga having 4096 colours? ), 16bit is 63336, 24bit is 16.7 million, and 30bit is 1 billion.

Usually we store the image in 3 parts, red, green and blue - so 30bit colour means there are 10bit for red, 10 for green and 10 for blue, compared to 8 for each in 24bit colour.

Sometimes we count alpha bits (as posted above), but Matrox was the first with true 10/10/10bit colour.

Is that of help?

P.

[thop]

So, an 8bit image is 256 colours, 12bit is 4096 (remember the Amiga having 4096 colours? ), 16bit is 63336, 24bit is 16.7 million, and 30bit is 1 billion.

not completely true. for a 1 billion color image you need an insanely big image with a resolution of 1mio x 1mio pixel
every pixel can only have one color. it is just that every pixel can choose from a variety of 1 billion colors instead of 'only' 16.7mio.

[Tempest]

Beep. The "american billion" is 1,000,000,000 (Parhelia's 2^30 is 1,073,741,824) so you'd "only" need a bitmap with dimensions of sqr(1,000,000,000) or ~31,623 pixels to exceed 1 billion colours. This is damn confusing for us Europeans

ps. 2^16 is actually 65,536.