Antialiasing broke on GF2?... 2x2, 3x3, 4x4 no difference

[RobsTV]

Just installed Nascar Heat, and ran the benchmark. Then tried all the different AA settings in my "more D3D" under display properties of Windows. Results didn't seem correct, so I ran Viper Racing benchmarks with all AA settings. Same sort of results. Final test was 3DMark 2000, which confirms something is not right. Below are the numbers.

Nascar Heat
No AA, 767MPix, 4061KTris,
2x2 AA, 192MPix, 4060Ktris
3x3 AA, 85MPix, 3136KTris
4x4 AA, 85MPix, 3084KTris

640x480=50.4 no AA
2x2 Low detail=34.4, 2x2 normal=50, all 3x3 and 4x4 scored 40fps.

800x600=50.3 no AA
2x2 Low detail=26.6, 2x2 normal=46, all 3x3 and 4x4 scored same 46fps.

1024x78=50.2 no AA
2x2 Low detail=18.7, 2x2 normal=31, all 3x3 and 4x4 scored 34fps.

1280x1024=50.4 no AA
unavailable with AA forced.

--------------------------------------

Viper racing scored same benchmark 47-49FPS at all resolutions and AA setttings, but I could easily see AA was indeed working.

---------------------------------------

3DMark 2000 scored 6850
2x2 AA low detail 2649
2x2 normal scored 2551
all other 3x3 and 4x4 settings scored same 2551


Perhaps it is a driver issue, or perhaps it is the mod I did to the card, but hopefully others can confirm or dispute these results, or explain how this can be. If it is an error, which AA setting is it defaulting too? Note that Nascar Heat does show a difference in throughput with 2x2 and 3x3. Both racing games looked great and ran smooth at all settings. If this is normal, does that mean I should use 4x4 regularly?

System specs that might have made a difference. Visiontek GF2 GTS. Resistor's moved to turn it into Quadro2 Pro. Flashed Bios to Elsa Gloria III Quadro2 Pro card. nVidia 6.47 drivers.

 

[BenSkywalker]

Hmmm, haven't heard this question in some time, I wrote up an article on a screw off site I setup once, let me see if it still exists....

D@mn, the whole site is down(Crosswinds) so I can't check.

Pretty much in summary it comes down to the fact that you can't exceed available RAM. How much that is, and exactly what resolutions will work with what FSAA settings is application dependant(texture loads and such varry memory loads).

The drivers are built to detect when available on board RAM is exceeded and "slide back" to the highest resolution allowed by available RAM. I ran over two hundred benches for the article in Direct3D alone, and found that there was no way you could tell in a game unless it had a bench(so you could see the FPS drop).

If you have a 64MB board you can run quite a bit higher then a 32MB offering, but your performance will be absolutely horrible of course.

As a general rule of thumb pretty much anything over 640x480 is limited to 2x2 ona 32MB board, to utilize 3x3 at even 800x600 means the scene must be rendered at 2400x1800.

Hope that clears it up

 

[RobsTV]

Thanks, it is beginning to make some sense. Tests were done on 32meg card.

So what does it default to if you do not have enough memory for the requested setting? 2x2 normal isn't consistant enough to be the answer, according to these tests? Or does it keep selected rate, (I.E.3x3), and drop down from say 1024x768 to 640x480? Also, why does 2x2 normal score higher FPS than 2x2 low detail, in the games tested?

After searching popular test and review sites, there is really no information that addresses this that I can find. Why put in 10 FSAA options, when you are basically limited to using only 2 of them, and that's only at low resolutions?? What card can do 800x600 using 4x4, or 3200x2400 working resolution?? Or do I have this wrong?

 

[BenSkywalker]

It drops FSAA down to the highest supported available option. For instance if you are running a texture heavy game and set the resolution at 1280x960 then you will only be running 1x2 FSAA no matter what the slider is set for. Up it to 1600x1200 and FSAA is off, no matter what it is set to.

Keeping the slider at 4x4 ensures that you will always be using the highest available FSAA setting. In actuality, many games won't run 4x4 in anything above 320x240(no joke).

"Also, why does 2x2 normal score higher FPS than 2x2 low detail, in the games tested?"

Your Heat scores have me scratching my head, I'd need a few hours running through some tests to figure out WTF was going on there, your 3DMark results should be indicative of what you should see though.

 

[RobsTV]

And the final question is, then why bother?
Using Nascar Heat for example, playing at 1280x1024 50fps, no FSAA, looks much better than playing at 640x480 40fps and FSAA, (IMHO). So is FSAA currently just a lot of smoke?

 

[BenSkywalker]

It is a preference thing. For games that don't use any mip mapping, such as many simulations, the texutre aliasing is absolutely horrible and FSAA makes a big impact. In those cases I think it is useful.

I definately take higher res instead of current FSAA over 90% of the time, though upcoming MSAA combined with high tap anisotropic should perform significantly better and avoid the blurring issues with current implementations.