FSAA + PS2.0 test in 3DMark03 = artifacts

[Ante P]

Could anyone try this out.
Just run the Pixel Shader 2.0 test in 3DMark03 (preferebly in Image Quality mode to grab a screen in the Pro version).

It doesn't matter if I use MSAA or SSAA, I still get these annoying white dot artifacts.

Could someone verify this and test if it also does this on non-ATi cards? Thanks.

 

[Tahir2]

Running the game Combat Flight Simulator 3 with 4x Anti-Aliasing enable results in texture corruption being displayed. This issue occurs under Windows XP with the RADEONâ„¢ 8500. The current workaround is not to use Anti-Aliasing
Enabling extended desktop mode and setting the display resolution to 1024x768 32bpp results in the game Aquanox and Star Trek Armada 2 failing to respond. This issue occurs under Windows XP with the RADEONâ„¢ 9000 installed. The current workaround is to either disable extended desktop mode, or trying a different resolution
3Dmark2003 GT4 runs a little slower than expected on certain chipsets where AGP writes are disabled. A future CATALYSTâ„¢ will resolve this and provide a higher benchmark score
3DMark2003 has some minor corruption on the RADEONâ„¢ 9500 and RADEONâ„¢ 9700 series cards when running in Anti-Aliasing modes. This however does not impact scores Star Wars Jedi Knight 2 may hang when using High Quality setting in the video quality selector. This is a known issue for the RADEONâ„¢ 7500. Try lowering your video quality slider
Myth III may crash when enabling TRUFORM for the RADEONâ„¢ 9500 and RADEONâ„¢ 9700 series cards. The current workaround is to not use TRUFORM at the current time

 

[Ante P]

Ahh spanks.

 

[ET]

I love "this however does not impact scores." Just shows where the priorities lie.

 

[andypski]

It doesn't matter if I use MSAA or SSAA, I still get these annoying white dot artifacts.

How are you choosing SSAA?

 

[Ante P]

It doesn't matter if I use MSAA or SSAA, I still get these annoying white dot artifacts.

How are you choosing SSAA?

non maskable in the "pixel processing" options of 3dmark03 (not "combinable" with MSAA like some people said though )

 

[Ostsol]

I love "this however does not impact scores." Just shows where the priorities lie.

Um. . . That quote can be read as suggesting that the rendering errors do not affect scores negatively or positively. Besides, it's not like the recent beta NVidia drivers that took out entire effects in GT1. . . I'm sure that had quite the effect on scores.

 

[andypski]

non maskable in the "pixel processing" options of 3dmark03 (not "combinable" with MSAA like some people said though )

That will still not be SSAA, afaik it'll just be non-maskable MSAA. It just equates to the D3DMULTISAMPLE_NONMASKABLE setting in the API. Nothing about super-sampling here...

I believe that if you select non-maskable AA on a GF3/4 etc it'll also still just be MSAA, and then the selectable quality levels will give you 2x, quincunx etc.

Anyway, I expect that the artifacts you're seeing would be the same on GF:FX or on a GF4 (if it could actually run the test). It would be interesting to get some confirmation, though.

 

[andypski]

Um. . . That quote can be read as suggesting that the rendering errors do not affect scores negatively or positively.

That's exactly what it means.

 

[Xmas]

That will still not be SSAA, afaik it'll just be non-maskable MSAA. It just equates to the D3DMULTISAMPLE_NONMASKABLE setting in the API. Nothing about super-sampling here...

Actually, the name is quite misleading, as the driver could chose to offer any kind of FSAA through this, not only multisampling.

 

[andypski]

That will still not be SSAA, afaik it'll just be non-maskable MSAA. It just equates to the D3DMULTISAMPLE_NONMASKABLE setting in the API. Nothing about super-sampling here...

Actually, the name is quite misleading, as the driver could chose to offer any kind of FSAA through this, not only multisampling.

True - I really should have just said it won't necessarily be SSAA. Of course, the back end can generate the multiple samples however it chooses.

 

[Dave Baumann]

Anyway, I expect that the artifacts you're seeing would be the same on GF:FX or on a GF4 (if it could actually run the test). It would be interesting to get some confirmation, though.

I'll have a look over the weekend

 

[Ante P]

non maskable in the "pixel processing" options of 3dmark03 (not "combinable" with MSAA like some people said though )

That will still not be SSAA, afaik it'll just be non-maskable MSAA. It just equates to the D3DMULTISAMPLE_NONMASKABLE setting in the API. Nothing about super-sampling here...

I believe that if you select non-maskable AA on a GF3/4 etc it'll also still just be MSAA, and then the selectable quality levels will give you 2x, quincunx etc.

Anyway, I expect that the artifacts you're seeing would be the same on GF:FX or on a GF4 (if it could actually run the test). It would be interesting to get some confirmation, though.

I just took for granted that it was SSAA due to the large performance hit and the fact that the HUD got all messed up.

 

[andypski]

I just took for granted that it was SSAA due to the large performance hit and the fact that the HUD got all messed up.

Hmm - the HUD doesn't get affected when I run it with AA. I doubt that I'm using the same driver version as you (in fact I can pretty much guarantee it ), but I don't think I've seen any problems with the 3DMark03 HUD even with previous driver versions.

[edit] Actually the HUD rendering does change slightly with AA on. [/edit]

 

[Ante P]

I just took for granted that it was SSAA due to the large performance hit and the fact that the HUD got all messed up.

Hmm - the HUD doesn't get affected when I run it with AA. I doubt that I'm using the same driver version as you (in fact I can pretty much guarantee it ), but I don't think I've seen any problems with the 3DMark03 HUD even with previous driver versions.

[edit] Actually the HUD rendering does change slightly with AA on. [/edit]

Just to clarify, the HUD doesn't get "messed up", it just looks like "2D" stuff does when you use SSAA. Also the performance hit seems much too large compared to the level of AA that is applied.
And this is of course only when using "non maskable" in the options. Not when using "normal" AA.