Why didn't nvidia and ati remove HW support for lower precision modes?

[Anarchist420]

Are 16 and 24 bit z-buffer modes really necessary? I was thinking all that was necessary to have hardware support for was D32FS32F.

Also, why didn't they get rid of the RGB10A2, R11G11B10, and RGBA8?

FP16, FP24, and FP32 are enough, aren't they? I know that using FP16 for a DX7 game would be a lot slower, but DX11 HW has the horsepower for it, right?

They know better than I do though, so maybe it really is necessary to have the lower precision modes.

 

[brandonb]

The same reason why we have serial, parallel ports, floppy disk drive, ps/2 connections, bios, PATA(IDE) connectors, stereo speakers (when 5.1 is newer and better)...

It's there just because...

 

[digitaldurandal]

backwards compatibility and GPU computing? I am sure there are programs still used that use all of the above methods, however I am not an expert to point them out.

 

[Absolution75]

Probably because the amount of die space is insignificant compared to how big of problems it could create.

 

[Anarchist420]

What problems would it create?

 

[digitaldurandal]

What problems would it create?

Programs not functioning properly.

 

[JimmiG]

stereo speakers (when 5.1 is newer and better)...

Yes, because I never go to a concert unless the band is standing in a circle with me in the center

 

[Absolution75]

lol vbulletin

 

[pw38]

....stereo speakers (when 5.1 is newer and better)...

It's there just because...

Huh? According to who? And for what?

 

[Anarchist420]

Programs not functioning properly.

How/why would they not function properly? I would imagine that a higher precision format would always be a perfect replacement for a lower one.

 

[boran]

ah, but that means your driver has to do the translation calls.
And what if you have a game that uses the imperfect results as a feature.

Dont think that does not happen, go and read the old new thing blog from raymond chen, one mans defect is another mens workaround.

 

[taltamir]

stereo speakers (when 5.1 is newer and better)...

5.1 isn't newer and better, 5.1 is just more speakers in a circle around you. (5.# = 1 central speaker, 1 left front speaker, 1 left back speaker, 1 right front speaker, 1 right back speaker; #.1 = 1 subwoofer) while stereo is just using 2.0 speakers (left front, right front).
Using two speakers is 2/5 the price of using five (of identical quality, which is what you should use because its better to have 2 high quality speakers than 5 crappy ones), and since you cut out subwoofer too you end up paying even less than that.
Stereo is more then good enough for the price and when working on a tight budget you can get better quality sound overall by going stereo (2x 50$ speakers is better than a 20$ subwoofer and 5x16$ speakers).
Besides all of that, most people don't even have a dedicated room with enough clearance and space to properly place back speakers.

 

[Ross Ridge]

Are 16 and 24 bit z-buffer modes really necessary? I was thinking all that was necessary to have hardware support for was D32FS32F.

Also, why didn't they get rid of the RGB10A2, R11G11B10, and RGBA8?

FP16, FP24, and FP32 are enough, aren't they? I know that using FP16 for a DX7 game would be a lot slower, but DX11 HW has the horsepower for it, right?

They know better than I do though, so maybe it really is necessary to have the lower precision modes.

Games avoid using the floating point formats except in the limited circumstances where they can benefit from the extended range they offer. 24-bit RGB, with 8-bits per component, offers enough precision for most purposes. That's more precision than most LCDs are capable of displaying. So they'll use floating point render targets to do HDR lighting, but most everything else will be in the same formats as have always been used. It would just be too slow and use way too much memory to have everything stored and rendered using floating point formats.

 

[Ben90]

ah, but that means your driver has to do the translation calls.
And what if you have a game that uses the imperfect results as a feature.

Dont think that does not happen, go and read the old new thing blog from raymond chen, one mans defect is another mens workaround.

Yep, it could mean the difference between an intentional .01 or divide by zero.

And I laugh at the laptops that have 5.1 speakers, how the hell are you supposed to take advantage of that? Put your nose on the "G" key?

 

[Anarchist420]

thanks to everyone for the replies. what about the day when FX function disappears and everything, other than textures, is emulated? Will old games be screwed then?

Or can FP32 shaders emulate 24 bit FX Z-buffer without having errors in games?

 

[Ross Ridge]

thanks to everyone for the replies. what about the day when FX function disappears and everything, other than textures, is emulated? Will old games be screwed then?

Or can FP32 shaders emulate 24 bit FX Z-buffer without having errors in games?

The fixed-function pipeline disappeared from video cards long ago.

A 32-bit floating point Z buffer can't be used to emulate a 24-bit integer Z buffer without causing compatibility issues. It could just easily break a modern DirectX 10 game that uses a 24-bit Z buffer as an old Direct3D 7 one.