32Bits Z-buffer

[Humus]

Need to remember you can get higher precision in the output pixels than the vertices because the pixels use interpolated values from the vertices.

Yes, but that's usually not relevant. We're not interested in "smooth" depth buffer values, as one would be in for instance the texture filtering case, where the precision of the stored values often isn't that important but rather at what precision they are interpolated. But for depth testing it's not so much the interpolated precision that matters but the computed positions of the output vertices. Good interpolated precision will for instance not help nearly coplanar surfaces if the vertices you're interpolating between have poor precision.

 

[Hyp-X]

Need to remember you can get higher precision in the output pixels than the vertices because the pixels use interpolated values from the vertices.

If the vertex Z has 1/2^24 error that the interpolated values will also have 1/2^24 error even if you calculate them with 128-bit arithmetic.
This is the same thing Humus is saying: Interpolation can't help you.

 

[HAL-10K]

Fact is that a higher precision depth buffer is _BADLY_ needed because of a more and more increased range of view and reduced triangle sizes in games.

It would increase performance for games which have to do ugly work-around stuff like additional depth clears, mesh sorting, etc.
I don't get it why the (consumer) graphics hardware industry is still stuck with 10 year old precision there!

This did and does cost us a lot of headaches and performance in our game.

 

[Ozo]

Based on all your posts, it seems that none of the high-end GPUs support more than 24bits Z-Buffer (i.e. ATI, nVidia and 3DLabs recent cards).

Is there anything else available to improve depth-buffer precision from these graphics cards, like 24-bit floating point W-buffer or 24-bit complementary floating-point Z-Buffer? I know ATI only supports pure simple 24-bit integer Z-Buffer. Are nVidia and 3DLabs the same?

Ozo.

 

[Lazy8s]

If the portable market is applicable, I think MBX has something like a 32-bit floating-point depth buffer.

 

[Dave B(TotalVR)]

I would be very suprised if MBX did not have a floating point 32 bit z buffer with w buffer support as powervr tech has supported this since the apocolypse 3d, back in the days of the voodoo 1.

 

[Simon F]

If the portable market is applicable, I think MBX has something like a 32-bit floating-point depth buffer.

No, it won't be anywhere near that accurate - but I don't know the actual figure.

 

[HAL-10K]

W-buffering can not be used in OpenGL.
It would be just another ugly workaround anyway since a higher precision close to zNear is required.

It is unbelievable that we still have to deal with 10 year old 24Bit precision that causes unacceptable z-fighting at ~zNear*1000 already!


An internal precision of Float32 for fragment programs to do such "trendy" but near to useless things like HDR (the hype around it drives me nuts), but we still have a 24Bit depth buffer only?!

The game world is sick of all this "fight from room to room" crap.
The future game development has to break out of the shoe box environments and thus requires higher ranges of view and better depth buffer precision!

 

[Dave B(TotalVR)]

W-buffering can not be used in OpenGL.
It would be just another ugly workaround anyway since a higher precision close to zNear is required.

It is unbelievable that we still have to deal with 10 year old 24Bit precision that causes unacceptable z-fighting at ~zNear*1000 already!


An internal precision of Float32 for fragment programs to do such "trendy" but near to useless things like HDR (the hype around it drives me nuts), but we still have a 24Bit depth buffer only?!

The game world is sick of all this "fight from room to room" crap.
The future game development has to break out of the shoe box environments and thus requires higher ranges of view and better depth buffer precision!

Im still waiting for the fight from room to room to outside shooting in the windows of the room you were just in which you can see out of inside and all. going back to villagemark, weve all seen it. what if you bump the detail up to doom 3 / fear levels inside and out and recreate the whole village as a deathmatch level for some game, plus have high vantage points and all. DOD source comes close but you cant help but feel a lot of level engineering has been done to keep the visibility of other level areas down from all sorts of points.

 

[darkblu]

The future game development has to break out of the shoe box environments and thus requires higher ranges of view and better depth buffer precision!

rumor is, the next gen of game engines will render in multiple vertically/horizontally-split passes, so you can take the next logical step and partition your scene by depth too ; )

 

[HAL-10K]

rumor is, the next gen of game engines will render in multiple vertically/horizontally-split passes, so you can take the next logical step and partition your scene by depth too ; )

This would be a horrible fillrate eater because it destroys the z-occusion culling.

 

[yeilam]

ATI's hyper-Z

The advanced ATI cards (like X600) implement "hyper-Z technology". It amazingly solves Z fight artifacts. It overcomes the z precision of 32 bit by 16 bit. (ofcoarse 24 bit is available there and is powerful).

 

[Rolf N]

The advanced ATI cards (like X600) implement "hyper-Z technology". It amazingly solves Z fight artifacts. It overcomes the z precision of 32 bit by 16 bit. (ofcoarse 24 bit is available there and is powerful).

No.

 

[swaaye]

Try playing Bridge Commander, Klingon Academy, or Independence War II and tell me z-buffering is perfect lol.