Tom's GF4 anisotropic scores in his Parhelia review

I think this is time for some 'Mission Impossible Music'...
krank.gif
 
ben6 said:
Hrm after some fooling around with Geforce4s I've noticed that oftentimes, anisotropic filtering will be disabled when running a benchmark with AA+anisotropic filtering. Defaulting to a lower setting. Tried Quake3 last night , for example, 1280x1024 2x FSAA 32tap and 64tap scored exactly the same, using the Nvidia control panel. Using Rivatuner , however there was the expected % drop in performance going from 4x to 8x .
Click to expand...

Are you saying that in the benches discribed above the AF was disabled possibly? That would explain the performance increase they discussed.
 
Quake 3 - Max Quality - Demo Four - GeForce4 Ti 4600 - 1280x1024

Detonator 29.42 (latest official Detonator drivers)
Detonator 27.30 (beta version used in my GeForce4 Ti 4600 preview)

No AA - Trilinear - 180 vs 178
No AA - 2X Aniso - 142 vs 119
No AA - 4X Aniso - 121 vs 90

2X AA - Trilinear - 130 vs 130
2X AA - 2X Aniso - 113 vs 100
2X AA - 4X Aniso - 100 vs 82

I took screenshots using the 29.42 drivers and the effects of anisotropic filtering can certainly be seen.
 
Mike,

John was asking about 4xOGMS numbers at Tom's where the gap between Level2 Aniso and Level8 Aniso seems to small.
 
That still doesn't change the fact that noAA/aniso or 2xAA/aniso numbers are irrelevant.

Just because the numbers have been cross-referenced it doesn't mean that all of us were able to draw a satisfying conclusion. My little curious nature just wants to know why and the data so far is more contradicting than explanatory to me.

I've sent an e-mail off already; no idea though if and when I will get an answer.
 
Contradictory? In what way?

DaveB and I already showed pretty effectively that the difference between 2x and 8x aniso is very small when 4x FSAA is enabled.

One thing you have to remember is that anisotropic incurs a fillrate hit and a very small memory bandwidth hit, while AA incurs a memory bandwidth hit and almost no memory bandwidth hit. Thus, when they are used together, they reduce the performance hit of each other.
 
Chalnoth said:
Contradictory? In what way?
Click to expand...

What's contradictory is that the texture LOD has definitely been changed in the latest drivers, which should negatively impact performance. Texture aliasing has therefore increased. Users are also noticing that they are having difficulty discerning a difference between 4x and 8x aniso with these latest drivers. And in spite of this the performance drop switching from 2x to 8x has been severely decreased from earlier drivers.

I'll stop beating around the bush. It very much looks like Nvidia is lowering their aniso quality and yet are attempting to hide/mask this by negatively increasing the texture LOD. Benchmark scores go up, and the majority of reviewers will never notice the IQ differences.
 
What's even funnier John is that I don't get LOD offsets as Dave showed in his screenshots with 29.42, neither in performance nor quality mode.

Shoot me but I still have the weird feeling that the 29.4x drivers if no 3rd party tweaking utitlity is used flip automatically to performance mode, at least when 4xFSAA is enabled at Level 8 aniso.

In the meantime I'm having a hard time to keep performance and quality mode output apart, exept maybe from lightmaps in q3a with Level 8.

(Thanks Reverend for the effort).

Chalnoth,

(from Rev's short article above) Besides the fact that performance has increased overall throughout all anisotropic filtering levels in the 29.42s compared to the 27.51s, the percentage drop in performance going from one level of anisotropic filtering to the next has drastically been reduced in the 29.42s compared to 27.51s.
Click to expand...

If someone else's results - whoever that may be - contradict my own and what my eyes are able to detect then of course I'll continue asking. I'm entitled to doubt anyone and don't need to be an expert either.

It's not an "issue" to me, nor something strange. What I like to know is what the real difference between performance and quality mode is, to cut it short.

(Rev) Personally, I see no difference in terms of overall image quality or anisotropic filtering quality between the two driver sets.
Click to expand...
 
I know for one thing I like the 29.42 drivers better because I can now use 8x anisotropic filtering in higher resolution game play where as before that wasn't even possible :D. Quality wise I really don't see the difference except now it is better since I am at a higher resolution.
 
That sure is a solid difference in lod with the two drivers ( Dave's shot). Adding lod and taking off filtering to gain speed and the only feature that really caters to texture aliasing. Fans of ridding the screen of texture aliasing artefacts will not be too happy with this driver set, imho.
 
That's what it looks like here with 8 aniso and a normal config in SS:SE. Of course using the extreme addon pushes the texture quality slider to the right with 29.42. That's one detail I don't remember wether it happened with 28.xx drivers and SS:SE.

http://users.otenet.gr/~ailuros/8anisoDef.jpg

edit: I just tried to set aniso through the drivers, while leaving it disabled in SS:SE and vice versa. Apparently the game overrides any driver aniso setting, or there's something utterly wrong in my setup.
 
SirPauley,

Yes the LOD does change for each setting of anisotropic filtering. Here are 4 images that show this on my GF3 Ti200 with the 29.42 drivers. Only difference in the settings between the images is the setting of anisotropic filtering in the Advance Options menu. Mipmaps are colorized.

No aniso

2x anisotropic filtering

4x anisotropic filtering

8x anisotropic filtering

Ailuros, that is normal for Serious Sam, it controls anisotropic filtering in the Advance Options menu.
 
Reverend said:
I did a very quick article, using MOHAA.

http://www.beyond3d.com/reverend/gf4aniso/index.php
Click to expand...

Thanks for taking the time to do that analysis. I've looked at both sets of screenshots carefully and was able to spot two subtle differences.

The solid brown lines on either side of the bottom right door frame do not show up with the newer drivers. However the line continues to appear on the left and top left frames. My guess is that the lines shouldn't be there.

On the ceiling to the left of the light nearest to the player. Notice that the elongated U shaped bracket looks correct with the newer drivers.
 
Still the performance drop when using anisotropic filtering is rather large. In addition the performance drop in combination with MSAA is still significant on GF3's. Here are some graphs I did testing SSSE with only anisotropic filter setting changes and or MSAA 2x.

anisograph.gif


anisograph2.gif
 
You must log in or register to reply here.
Back
Top