WTF is up with nVidias Z-buffer?

[Ante P]

I've got CRAZY amounts of Z-buffer "fighting" in games that use a 16 bit Z-buffer on my Ti4600 that never were there on my 7200, 7500 or 8500.
(Even when not forcing a 24/32 bit Z-bufer on the Radeons.)

I find myself simply shooting at artifacts all the time when using long range weapons.

Omegadrive mentioned forcing a higher z-buffer precision on nVidia cards recently. But no tweaker I've seen has this possibility, so how can I fix this?

 

[Nagorak]

Bah! You shouldn't have bought an Ncrapia!

Hehe, just kidding.

I'm not sure forcing a higher z-buffer actually does anything (wouldn't apps programmed for a 16 bit buffer only take advantage of it up to 16 bits?). You might want to download the 30.something drivers because they are supposed to be a lot more stable than the 40.somethings. I don't have any Nvidia cards though, so I'm just going off what others have said.

 

[Ante P]

I've tried 30.82, 31.00, 31.40, 40.41, 40.52 and 40.71 and they're all the same in this aspect.

Luckilly I don't buy videocards, I get them for free.

Well anyways, it's not the games fault since they look good on the 3dfx, ATi and Matrox cards I've tried.
I don't really remeber seeing this problem on the GF2 or 3 so perhaps I need to go back to an even older driver.

Which one was the first to have full official and properly working support for GF4?

 

[K.I.L.E.R]

Yes there is also artifacts in games that use a higher z-buffer.
Anyone who has played Warcraft 3 and been on the 2nd or 3d human mission where the 2 kightless horses come back to Uther?
There seems to be white lines if you look at the sequence from scratch and carefully.

 

[Hyp-X]

If you want an nVidia card with precise Z-buffer, well the GF2(!) has it.
With GF3 the precision has gone done considerably.
The GF4 is actually better than the GF3 but still nowhere near the GF2.

I suspect it's the effect of nVidias lossless Z-buffer compression.

I've found it interesting that Radeon cards have used Z-buffer compression from the start yet there's no reduction in quality.

 

[Gollum]

You could try disabling the Z-Buffer compression (Rivatuner has this option) and see if it changes anything for you. Strangely enough I have never experienced any Z-Buffer artifacts with my Ti4200, but I am probably only playing a limited number of games right now that might be critical for showing Z-Buffer inaccuracy and exhibit problems.

I couldn't reproduce the Warcraft 3 thingy BTW, maybe I wasn't looking at the right places, but I didn't notice anything unusual at all in that sequence...

 

[Nagorak]

If you want an nVidia card with precise Z-buffer, well the GF2(!) has it.
With GF3 the precision has gone done considerably.
The GF4 is actually better than the GF3 but still nowhere near the GF2.

I suspect it's the effect of nVidias lossless Z-buffer compression.

I've found it interesting that Radeon cards have used Z-buffer compression from the start yet there's no reduction in quality.

Radeon cards have graphical glitches associated with Hierarchical-Z, so they aren't perfect either. Hierarchical-Z is the Z-occlusion aspect of Hyper-Z I believe, and it's silly because it has to be disabled all the time for certain games to function right. This is true on both the R8500 and original Radeon, maybe not for the R9700... It only affects certain games, like every single one using the Lithtech engine (sigh). But it's a fairly common problem/fix.

 

[Dave Baumann]

There has been talk before that Radeon (orig and 8500) may not even store a full level Z buffer when Hier Z is enabled. I can't actually verify this though.

 

[Ante P]

It only affects certain games, like every single one using the Lithtech engine (sigh). But it's a fairly common problem/fix.

I've heard of a few games and applications that suffer from this. Haven't seen any problems in any Lithtech based game.

Haven't experiences any problems myself that I can remember, though I think there was some driver which generelly caused artifacts due to a broken Hyper Z implementation.

 

[Ante P]

Well anyway I've tried running DoD with Page Flipping and Block Transfer. I Also tried forcing a 16 bit Z-buffer. Didn't help.

So I tried running it in D3D and disabling the Z-compression. Well performance took a bad hit but the problems remained.
Tried disabling 24 bit Z-buffer and problems still remained.

Does using nv1x emulation disable the Z-buffer "optimizations"?
Well I'll try it anyway.
I'd like to use SSAA anyways. (But when I tried using that emulation soem time ago it didn't work at all, is there anything except turning on the option in rivatuner that I need to do?)

 

[Ante P]

Well I tried nv1x emulation.
Sure enough SSAA work, but performance is crappy (worse than the 8500 Quality Smoothvision) and the quality (except for AA on vertical edges) is crap. Well I can't really blame 'em for this since it's not officially supported. Aniso is capped at 2x, any way to work around this?

Well anyway, it didn't do anything for the Z-buffer problems.

Well all that remains is going back to an older driver, so I ask again. Which was the frist driver to support GF4 that don't have a lot of "initial" problems so to speak.

 

[mat]

i remember some z-buffer artifacts in Serious sam (1 and 2).
but only on a few locations on some objects far away. (that was on my old GF2MX, havent tried it yet with my Ti4200)

 

[Randell]

the 23.11's and the 27.20's are meant to be good drivers according to my Gf3 owning friends (not sure if the 23.11's support the Gf4 though).

 

[Tahir2]

There has been talk before that Radeon (orig and 8500) may not even store a full level Z buffer when Hier Z is enabled. I can't actually verify this though.

If you enabled FSAA on the Radeon 8500 I heard it no longer uses any kind of Hyper Z compression... perhaps this would alleviate some of the issues mentioned here?

 

[Ante P]

There has been talk before that Radeon (orig and 8500) may not even store a full level Z buffer when Hier Z is enabled. I can't actually verify this though.

If you enabled FSAA on the Radeon 8500 I heard it no longer uses any kind of Hyper Z compression... perhaps this would alleviate some of the issues mentioned here?

According to the benchmarks I've seen and replicated on my own machine Hyper Z is disabled when running 3x FSA or more. 2x will not disable it. (I think Quality vs Performance also had some impact on the matter)

 

[Ante P]

i remember some z-buffer artifacts in Serious sam (1 and 2).
but only on a few locations on some objects far away. (that was on my old GF2MX, havent tried it yet with my Ti4200)

Yes but that's thanks to croteam, they use a 16 bit Z-buffer and in some of the more complex scenes objects in the distance gets really screwed up.

"Force 24 bit Z-buffer" doesn't work on the 8500 either so I think they force it somehow (dunno how though, I thought forcing a 24 bit Z-buffer simply disabled support for 16 bit, at least that's how it works in D3D.. oh haven't tried D3D and forcing a beter precision, but I'm too lazy to put my 8500 in again )

 

[christoph]

Ante P wrote:

Yes but that's thanks to croteam, they use a 16 bit Z-buffer and in some of the more complex scenes objects in the distance gets really screwed up.

you can change this in-game to 24bit z / 8bit stencil and the artefacts are gone

 

[Ante P]

Ante P wrote:

Yes but that's thanks to croteam, they use a 16 bit Z-buffer and in some of the more complex scenes objects in the distance gets really screwed up.

you can change this in-game to 24bit z / 8bit stencil and the artefacts are gone

huh???
not in video option and not in advanced rendering options
console commands or command lines?

 

[Nagorak]

It only affects certain games, like every single one using the Lithtech engine (sigh). But it's a fairly common problem/fix.

I've heard of a few games and applications that suffer from this. Haven't seen any problems in any Lithtech based game.

Haven't experiences any problems myself that I can remember, though I think there was some driver which generelly caused artifacts due to a broken Hyper Z implementation.

Have you played NOLF or AVP2? In both Lithtech engine games, the chrome on metal was totally messed up. Basically the lower texture would clip through the outer metallic texture and make it look very bizarre. It didn't make the games unplayable, but it was somewhat annoying. Also in Tribes 2 there was some pixel popping associated with Heirarchical-Z. Basically the structures in the game would have discolored pixels appear and disappear on them. Once again, it wasn't bad, but it was definitely noticeable. Disabling Hierarchical-z fixed the problem there too. I believe I've run into a few other problems with it, but those are the only two major ones that come to mind.

I really hope the R9700 H-Z routine has been shored up a little because I don't like having to "worry" about it causing problems.

 

[Foodman]

I use the 28.32s and have no problems in any game (Well, the shadows glitch in Ground Control now and again, but the game is pretty old). They just work.