Geforce FX Bilinear Anisotropic Filtering Question ??

Xmas said:
Ante P said:
Xmas said:
Joe DeFuria said:
I don't know what marketing guy came up with the aniso "terms", (Balanced and Aggressive), but they just plain suck. :!: Balanced in particular. Doesn't that imply some "happy medium" between high quality/low performance and low quality/high performance modes?
Yes, it does. But what's wrong about that?

Since there's no high quality mode it's a bit too vague, perhaps that's what he's implying?
Huh? There is, what do you think that 'Application' setting on that slider means?

It means that the driver will hand over controll to the application when it comes to AA, texture filtering, LOD etc.
So in most cases: no AA/AF.. if that means High Quality to you then by all means.
To me it looks like crap since I enjoy playing with max AA/AF. ;)

Balanced, Aggressive and Application actually worked in 41.09 or 41.07 or whatever driver it was.

Application left all options to the Application: disregarding what levels of FSAA/Aniso you chose manually. Simply put: "Application Preference."
Balanced is the normal mode that we're all used to, it doesn't do anything "funny".
Aggressive made the IQ look like doodoo, but some sort of of the features you "forced" through the drivers were still present unlike "Application" mode.

I assume the FX works with the same principles.
 
Thanks ram, that was more than revealing.

Nvidia advices to use the aggressive mode for comparison benchmarks to the R300 performance mode.

That sound more like it. I probably got the wrong impression that the "recommendation" was aggressive FX against quality R300. IMHO it would be best if reviewers would use all 4 modes in shootouts then (ATI performance/quality, NV balanced/aggressive), preferably accompanied with a small IQ investigation to let the final user decide what applies better to his needs.

Aggressive looks bad admittedly on the FX.

Xmas,

I thought I had your application and the server ram linked to is currently down. How about passing it over for download to some sites you trust? Nonetheless it'll be easier to find at all times.
 
Someone really needs to get some shots of what the mipmap transitions for the GeForce FX looks like in an actual game. The AF test apps shows a really strange pattern for the large lods.
 
Nvidia advices to use the aggressive mode for comparison benchmarks to the R300 performance mode.

It's impossible to really tell just by looking at the mip-levels, but based on them, I would say that the overall quality would go something like this (from worst to best):

NV Aggressive-->ATI Performance-->NV Balanced-->ATI Quality-->NV Application Pref.

Again, I'd like to know if the "high quality" (app preference) mode of the FX can be forced on, or if it's only available with applications that can specifically ask for anisotropic.
 
Colourless said:
Someone really needs to get some shots of what the mipmap transitions for the GeForce FX looks like in an actual game. The AF test apps shows a really strange pattern for the large lods.

I don't think it will be any different in a game, it is just much harder to get a scene where you can see all angles at the same time.
 
ram said:
As you can see, the Geforce FX "balanced" and "aggressive" modes seem to work angle dependent, but only on some mip maps. The R300 in contrast works angle dependent on all mip maps. The "old" application mode is the same as on the NV2x series, which is slow but offers the highest quality.
Thanks for the shots.
But btw, the odd "squarish artifact" looks also exists in the NV2x shot. It's just more faint. Gah, this is quite disappointing. All benchmarks should definitely be done with 'application settings' if possible. These definitely look like plain LOD adjustments.
 
The 'square' look, IMO could be due to a short cut with lod selection. To me (and i could be quite wrong about it), it 'almost' looks like they could be just using the one of the iterators when computing which mipmap to use (the 'biggest one'), rather than both. However once they become too small it then uses both.

With large lods with would generally be ok, except you'll get some blurring on the edges of the screen, and this may or may not be noticable, but with smaller levels, you would possibly have gotten a substantial amount of blurring, which is totally not what you want with AF.

I would expect an image of the ground in a game might have a rather severe ^ shape (with straight lines) instead of a nice curve.

Also, it would appear that it is doing trilinear filtering on the GFFX, but IMO, it looks far from optimal. There is some minor blending between lods (look at red->green transition).
 
Xmas said:
Ailuros said:
Xmas,

I thought I had your application and the server ram linked to is currently down. How about passing it over for download to some sites you trust? Nonetheless it'll be easier to find at all times.

It's available here: http://www.beyond3d.com/misc/Anisotest/

Thanks I got it from your server. I was stupid enough to try to fetch it with a d/l manager heh.

By the way there isn't a difference (in 41.09) on NV25 between the three modes with your application, which probably means that the balanced/aggressive optimisations are operational only on NV30. On NV25/openGL it's still application= performance mode Rivatuner, balanced/aggressive= quality mode Rivatuner.

In fact setting the slider anywhere outside of "application" in the driver panel, makes Rivatuner report an "invalid" anisotropic level driver extension (with or w/o Unwinder's AnisoBooster patch).
 
Colourless said:
The 'square' look, IMO could be due to a short cut with lod selection. To me (and i could be quite wrong about it), it 'almost' looks like they could be just using the one of the iterators when computing which mipmap to use (the 'biggest one'), rather than both. However once they become too small it then uses both.
I don't think so. Looking at it on my GeForce4, when looking at a flat surface, it looks more hyperbolic. This could be accomplished by subtracting the squares under the square root instead of adding, if I remember my conic sections correctly. I'm not entirely sure why this was done in the first place, and it's only done for surfaces at lower angles. When the surface is highly-angled, the odd MIP maps disappear.
 
Ante P said:
It means that the driver will hand over controll to the application when it comes to AA, texture filtering, LOD etc.
By the way, at least with the 41.03 drivers that I'm using, it still forces anisotropic when "application" is selected.
 
Application = Off or let the application choose in the ATI R200/R300 Control Panel.

Might help if you experience and problems with textures or AA not working correctly.
 
driver2.gif


Why does the slider work backwards ?? So if we slide the slider to aggressive it actually is lowering the anisotropy ?? If we leave it balanced its less than a Geforce 4's currently now, then move it to Application (to the left) actually improves it more by matching a Geforce 4 ??

Not exactly user friendly is it.
 
Doomtrooper said:
driver2.gif


Why does the slider work backwards ?? So if we slide the slider to aggressive it actually is lowering the anisotropy ?? If we leave it balanced its less than a Geforce 4's currently now, then move it to Application (to the left) actually improves it more by matching a Geforce 4 ??

Not exactly user friendly is it.

I dunno, there's no clear word here.
But when I tried this slider in some older driver this is what it did:

1. Application: ignores all your settings and uses the game defaults. (ie LOD, Aniso, FSAA, Z-buffer format etc. everything is controlled by the application)
2. Balanced: the normal mode we're all used to on nVidia products.
3. Aggressive: crappy IQ better performance.
 
I can't see any difference neither in performance or IQ in games or Xmas' AnisoTester with "aggressive", 41.09 on NV25. On the other hand I'm actually glad I don't :D
 
Doomtrooper said:
Why does the slider work backwards ?? So if we slide the slider to aggressive it actually is lowering the anisotropy ??

How is it backwards? Perhaps you didn't read the title of the slider, it says "Performance" not "Filtering Quality", thus logically moving it to the right should increase performance, which it does.
 
ahhhh no...

Both sliders below move to the right and increase quality...the top one has to be moved to Application (left) for quality
eek13.gif
 
Back
Top