FYI - Aniso settings are a completely different kettle of fish. And we'll look at the separately, for this comparison leave Aniso out of the equation.
Non AF IQ Comparison
- Thread starter Dave Baumann
- Start date
For the actual poll question, it seems pretty obvious to me...but for the discussion that has followed...I don't think a poll can work for more comparisons (and, actually, I think polls would disadvantage the GF FX since it can't be evaluated in motion), and I think there is a problem with all these "pretty mip map color" comparisons.
Hey, here I am sharing my opinion again (so rare, right?)...I'll use ~~ for the "depending on circumstances" evaluations as far as I recall from images I've seen over time, and I'm also presuming base texture LOD is the same (For FX App and 9700 Q, that seems to be the same as much as I can see through the coloring).
Trilinear (well, err...I mean "No aniso", since some of these aren't trilinear at all)
Pretty mip colors: 9700 Q = FX App > FX Bal > 9700 P >> FX Agg
Texture filtering: 9700 Q = FX App > FX Bal ~~ 9700 P >> FX Agg
FX Bal ~~ 9700 P: I'm not sure how significant the impact on the texture detail adjustments for FX Bal are on the image, but if there was no difference (though there does seem to be some sort of difference based on the screenshots), I'd have to presume that some blending, no matter how limited, between mip level transitions, would be better than none, so the FX Bal mode should offer advantages to offset the disadvantages in texture detail in comparison to 9700 P mode.
Aniso
Pretty mip colors: FX App > 9700 Q > FX Bal > 9700 P >> FX Agg
Texture filtering: FX App ~~ 9700 Q > FX Bal ~~ 9700 P >> FX Agg
For screen shots, AFAIK the 9700 wins for the above "~~" cases. In motion, well, I haven't compared personally and the first hand evaluations seem to go either way, and it might change for different games.
Hey, here I am sharing my opinion again (so rare, right?)...I'll use ~~ for the "depending on circumstances" evaluations as far as I recall from images I've seen over time, and I'm also presuming base texture LOD is the same (For FX App and 9700 Q, that seems to be the same as much as I can see through the coloring).
Trilinear (well, err...I mean "No aniso", since some of these aren't trilinear at all)
Pretty mip colors: 9700 Q = FX App > FX Bal > 9700 P >> FX Agg
Texture filtering: 9700 Q = FX App > FX Bal ~~ 9700 P >> FX Agg
FX Bal ~~ 9700 P: I'm not sure how significant the impact on the texture detail adjustments for FX Bal are on the image, but if there was no difference (though there does seem to be some sort of difference based on the screenshots), I'd have to presume that some blending, no matter how limited, between mip level transitions, would be better than none, so the FX Bal mode should offer advantages to offset the disadvantages in texture detail in comparison to 9700 P mode.
Aniso
Pretty mip colors: FX App > 9700 Q > FX Bal > 9700 P >> FX Agg
Texture filtering: FX App ~~ 9700 Q > FX Bal ~~ 9700 P >> FX Agg
For screen shots, AFAIK the 9700 wins for the above "~~" cases. In motion, well, I haven't compared personally and the first hand evaluations seem to go either way, and it might change for different games.
Joe DeFuria
Legend
Pretty obvious to me that Application is much more comparable to ATI's trilinear.
Brent said:
woops i missed that, i can't say i have seen the fx in person so i am not the best person to say. however, from all the reviews i have seen most notablity the tests Dave did here; i it seems clear to me that balanced is inferior quality.
and back to the topic; i know Luminescent just made i simple mistake but who are the other two vodes for ballanced, and why?
Hrm. I did what you suggested and I'd say that the balanced FX one looks better than the R300. (Shows more detail).
I'd have to see it in motion to know how well its working, though.
I'd have to see it in motion to know how well its working, though.
OpenGL guy
Veteran
I see what Russ and Hyp-x are saying about the ground, but I'm not convinced there's any real difference. Note that the left wall is a very high frequency texture and the R300 and GeForce FX application setting are giving pretty much identical results there. I wonder if the difference in the ground is just a result of the jpeg compression, especially considering that the mipmap transitions are happening at the same location.RussSchultz said:
Hyp-X/Russ:
Hrm, after seeing the comparison myself, the FX application setting seems to show more texture detail, though it predominantly seems to be happening in the red/pink bands. If you look at the yellow band it's much more subtle.
Dave: Would it be possible to get the images in PNG format, and to have another set of images that don't have the colorized mipmap regions for all three modes?
Nite_Hawk
Hrm, after seeing the comparison myself, the FX application setting seems to show more texture detail, though it predominantly seems to be happening in the red/pink bands. If you look at the yellow band it's much more subtle.
Dave: Would it be possible to get the images in PNG format, and to have another set of images that don't have the colorized mipmap regions for all three modes?
Nite_Hawk
Well based on the pretty mipmap colors, app is obviously closer than bal to qual, although I think app may have a slightly larger region of blending than does qual.
I'd still like to study pics (although motion would be better) without the mipmap coloring to see if this "2.5linear" filtering has any real effect on IQ. 2.5linear could be a brilliant idea, or it could be barely better than bilinear; at this point, I have no sense either way.
Finally, as to Hyp-X's argument that R300's texture quality appears worse in any case: looking at the stone texture at the bottom of the pics, it seems to me that FX app is very slightly but noticeably more detailed than FX bal. R300 qual is lighter than both of them; perhaps a more negative way to say this is "more washed-out". But I can't tell if it actually shows less detail. Maybe with an uncompressed pic...or maybe my eyes just aren't trained well enough.
(edit: oh, up in the red band! Yeah, R300 seems to be missing detail there.)
I'd still like to study pics (although motion would be better) without the mipmap coloring to see if this "2.5linear" filtering has any real effect on IQ. 2.5linear could be a brilliant idea, or it could be barely better than bilinear; at this point, I have no sense either way.
Finally, as to Hyp-X's argument that R300's texture quality appears worse in any case: looking at the stone texture at the bottom of the pics, it seems to me that FX app is very slightly but noticeably more detailed than FX bal. R300 qual is lighter than both of them; perhaps a more negative way to say this is "more washed-out". But I can't tell if it actually shows less detail. Maybe with an uncompressed pic...or maybe my eyes just aren't trained well enough.
(edit: oh, up in the red band! Yeah, R300 seems to be missing detail there.)
You using high quality or quality setting for texture pref on the ati shots? default for ogl is quality while its high quality for d3d.
I ask this because with a quick few tests on my 8500 colour mip levels (at least in the aniso test app) don't seem to be telling the entire truth below high quality. With colour mip levels on the transitions are smooth, turn them off and you get the (erm) 'double bilinear' effect, the one that a few sites mentioned at the 8500s release saying it didn't do proper trilinear. It's easily noticable by using +3 or more mip bias. Does this apply to the r300 chips too?
I ask this because with a quick few tests on my 8500 colour mip levels (at least in the aniso test app) don't seem to be telling the entire truth below high quality. With colour mip levels on the transitions are smooth, turn them off and you get the (erm) 'double bilinear' effect, the one that a few sites mentioned at the 8500s release saying it didn't do proper trilinear. It's easily noticable by using +3 or more mip bias. Does this apply to the r300 chips too?
voted for application, but after reading some of the comments. Im suprised to see that balanced seemed to have "clearer" textures around the red band. Trying to figure out why.
later,
later,
I compared the trilinear shots without the colored mip bands of Dave's 9800Pro review with the 5800Ultra preview and the texture detail seems to be similar. So the much more visible difference that HypX reported (and that I notice too) in the original colored shots of this thread could really be jpg artifacts.
Lossless compression pictures would be indeed nice.
Lossless compression pictures would be indeed nice.
