My R9700 experince and Aniso still "flawd"?

[Nagorak]

After an hour or so I remember I had put it in the 9700 box when I made the swap and I chucked that last week. Fortunately I had removed the card before I tossed the 9700 box in the trash and found it hiding in the cabinet in my printer return (whew!)

Ouch, that would have been BAD!

 

[darkblu]

Does the Radeon9700 anisotropic filtering algorithm determine anisotropy per triangle/polygon?

I don't know, but I don't think it matters... the pixel shape is the same no matter if you look at a single pixel or the whole triangle.

???
do you mind elaborating on the above statement of yours?

 

[OpenGL guy]

Does the Radeon9700 anisotropic filtering algorithm determine anisotropy per triangle/polygon?

I don't know, but I don't think it matters... the pixel shape is the same no matter if you look at a single pixel or the whole triangle.

???
do you mind elaborating on the above statement of yours?

What's important is du/dx, dv/dx, du/dy and dv/dy. They don't depend on the polygon, really, just the variation of the texture coords across the pixels. For example, the same polygon can have much different u,v's mapped to it so that will affect the results.

 

[darkblu]

Does the Radeon9700 anisotropic filtering algorithm determine anisotropy per triangle/polygon?

I don't know, but I don't think it matters... the pixel shape is the same no matter if you look at a single pixel or the whole triangle.

???
do you mind elaborating on the above statement of yours?

What's important is du/dx, dv/dx, du/dy and dv/dy. They don't depend on the polygon, really, just the variation of the texture coords across the pixels. For example, the same polygon can have much different u,v's mapped to it so that will affect the results.

i still don't get in what aspect the pixel shape is the same. the variation of texture coords across each pixel does depend on the polygon's orientation in eye/screen space. there's no such thing as single anisotropy degree per polygon (and this is how i interpret your 'pixel shape is the same' phrase) unless the polygon is viewplane co-planar, which is the sole exception.

 

[tEd]

@donald

i'm just curious , where did you change the aniso settings , outside the game(control panel,rivatuner etc) or in game (advanced rendering options)?

I suspect that your radeon_8x gf4_4x aniso are actually 16x and 8x aniso.

aniso settings in the driver control panel does nothing in ss2 for me.

 

[Bigus Dickus]

My comments:

I can spot no difference between the 9700's 8x and 16x quality, nor can I spot a difference between the GF4's 4x or 8x quality, whether in rotated or non rotated shots. Were the settings actually applied?

Assuming the settings were applied, the 9700's 8x looks better than the GF4's 8x on level surfaces.

The 9700's 8x looks better than the GF4's 8x on 45 degree rotated surfaces.

The GF4's 8x looks better than the 9700's 8x on 22 degree rotated surfaces.


These are my objective observations... I'll refrain from making a subjective opinion on which AF implementation overall looks better until more information on the actual AF settings applied is given.

 

[Joe DeFuria]

Donald...thanks much for doing that work!

And I pretty much concur with Bigus on everything he said. Though some of the shots quality is so close, it's hard to tell if it's a difference in the quality, or just a consequence of it being impossible to take shots in the exact same place / rotation.

On the level floor, 9700 is clearly sharper.

On the 45 Degree, I'd rate them about the same, or a very slightly sharper on the 9700.

On the 22 Degree, slightly sharper on the 4200.

 

[darkblu]

Bigus,

my observations are the same as yours. funny thing is, though, that not only at 8x but also at 16x the final 1-2 cm from the ~22 degr-rolled floor on the 9700 still look worse than same floor span on the gf's at 8x.
now, i may be wrong here, but it doesn't seem to me that 8x aniso covers the exibited aniso for the final few cm's of the floor..

 

[Donald]

Thanks guys and no problem on the work

Just to clarify...I actually took 2 sets of shots because I too saw that it looked like the settings were not being applied on the cards...especially the 4200 4x to 8x shots. So I double checked Rivatuner, rebooted, triple checked Rivatuner, triple checked game settings again and then took them again...same result. Now I can sit here and speculate as to what the cause is but truth be told I don't know. It could be the app, but my guess is that it's something in the drivers. It's just an opinion, but that's where I stand until it's proven otherwise.

I made sure that trilinear was applied in game on both settings (there is a bug in 9700 drivers that keeps the quality aniso working unless trilinear is selected in game...fixed in new drivers though). Otherwise the 9700 will back off to bilinear + aniso even if you have quality selected (that info from HardOCP). You are probably aware that there is also a bug in the 9700 drivers where if you select a new AA of AF mode you either have to change rez or reboot to apply settings. I rebooted between each change to insure that it was sticking (also fixed in new drivers).

I agree with you guys on what is better where. Now I have a better perspective since I can sit there and watch both work, be it stills or in motion. I agree that the 9700 has a better level and 45 degree, but the 22 is definitely a bit worse than the 4200. If you are sitting there watching that floor go back and forth without moving then you can clearly see the spots in question..uh..."get fuzzy"...and then clear back up again as it approaches the 45 and 0 degree planes. The Ti does not exhibit this behavior anywhere near the degree the 9700 does. The Ti in fact almost makes it look perfect when it is moving like that. The lines all seem to behave in a very believable manner. I was surprised to see the 9700 exhibit this behavior. Ok, I'm making it sound worse than it is. I played around with those shots in that room the first day and I did not notice any "shifting" going on. Only when I sat there and actually looked for something specific did it make itself apparent. It is NOT distracting, nor does it detract in any way from the overall image quality IMO. Well unless you are trying to find something wrong with it.

But I still contend that in motion it's REALLY hard to spot any discernible differences on either card. I think the 9700 is better because it actually applies the filtering to a point farther forward that the Ti does, especially in 16x. I haven't found a room long enough to figure out where 16x becomes blurry I think that these tests represent an extreme minority in real world gaming. Sure it's fun to test and compare, but how often do you get to see something like this? I have a ton of FPS games and SS is the only one without level floors

Bottom line is that the average dude, heck even the above average geek, that buys this card and plays games will NEVER notice anything but pure and wonderful IQ with the 9700. As well you should when you spend that kind of money for a video card.

 

[Sabastian]

Excellent Donald, we can compare your shots with ones done in this IQ review @ hardocp for a larger reference. http://www.hardocp.com/article.html?art=MzQ0

 

[Doomtrooper]

So there we have it folks, after much debate...and thanks to Wipeouts excellent work...this case is officially closed....I hope.

Thx Donald for all those screen shots..

 

[Bigus Dickus]

It could be that the room isn't long enough to spot the difference between the 9700's 8x and 16x on the level floor, and that when rotated they both default to the same mipmap distances, but you would think the 22 degree rotated should be a bit sharper with 16x than with 8x. That just doesn't seem to be the case.

Perhaps ATI's 16x and 8x look identical, much like the GF4's 8x and 4x are identical (there's little doubt there now). A longer room is probably needed to check for the difference in 16x and 8x.

Donald... is there any spot in SS where a repeating texture (similar to the floor) goes way off in the distance, like an outdoor scene perhaps? It would be nice to know if 16x and 8x on the 9700 look the same on level surfaces, or just when rotated.

(I also noticed in HardOCP's comparisons that I couldn't see a difference in 16x and 8x, but there were obvious distance limitations with the scenes he chose, so it was really useless).

 

[Donald]

There are some pretty massive outdoor areas with a lot of stones going up to temples and such. I will have to give that first level a good run through tonight and see if I can find something that stretches off into the horizon and snap off some shots.

 

[KimB]

Perhaps ATI's 16x and 8x look identical, much like the GF4's 8x and 4x are identical (there's little doubt there now). A longer room is probably needed to check for the difference in 16x and 8x.

No, the GF4's 8x and 4x are not identical. Unfortunately, they do look identical under Direct3D if the driver's aniso settings are used (meaning it's not properly applying anisotropic), but the Rivatuner/OpenGL settings are applied properly, and there is most certainly a difference.

Additionally, those shots should most certainly be good enough to tell the difference between 4-degree, 8-degree, and 16-degree anisotropic. I just think that the settings are not being applied properly. My suspicion is that 8x is always being applied in the GF shots, while 16x is always being applied in the Radeon shots.

 

[OpenGL guy]

i still don't get in what aspect the pixel shape is the same. the variation of texture coords across each pixel does depend on the polygon's orientation in eye/screen space. there's no such thing as single anisotropy degree per polygon (and this is how i interpret your 'pixel shape is the same' phrase) unless the polygon is viewplane co-planar, which is the sole exception.

Think about it this way:
- Take a simple rectangle with coords (0,0), (0,100), (100,0), (100,100)
- Apply a texture with tex coords (0,0), (0,1), (1,0), (1,1) respectively
- Now take the same rectangle with tex coords (0,0), (0, 0.5), (1, 0) (1, 0.5).

What happened? Same polygon, different texture coords, different pixel shape.

Edited to fix mistake in tex coords.

 

[Bigus Dickus]

No, the GF4's 8x and 4x are not identical. Unfortunately, they do look identical under Direct3D if the driver's aniso settings are used (meaning it's not properly applying anisotropic), but the Rivatuner/OpenGL settings are applied properly, and there is most certainly a difference.

Additionally, those shots should most certainly be good enough to tell the difference between 4-degree, 8-degree, and 16-degree anisotropic. I just think that the settings are not being applied properly. My suspicion is that 8x is always being applied in the GF shots, while 16x is always being applied in the Radeon shots.

Perhaps, but when the man says he used RivaTuner, Rebooted the 9700 between settings, and monkeyed extensively with settings in both the driver and in-game... double and triple checking his results, I don't see any other option than to take him at his word.

I don't take calling people a liar very lightly, and I don't have the hardware myself to confirm or disprove the results.

Maybe more screenshots in other games could help clarify things, but this thread after all was about investigating the 9700's behavior in rotated plane AF situations, and with Donald's work I think that has been answered.

 

[Bjorn]

..
On the level floor, 9700 is clearly sharper.

On the 45 Degree, I'd rate them about the same, or a very slightly sharper on the 9700.

On the 22 Degree, slightly sharper on the 4200.

Pretty much my observations also.

 

[KimB]

Perhaps, but when the man says he used RivaTuner, Rebooted the 9700 between settings, and monkeyed extensively with settings in both the driver and in-game... double and triple checking his results, I don't see any other option than to take him at his word.

I don't take calling people a liar very lightly, and I don't have the hardware myself to confirm or disprove the results.

Application settings for Anisotropic (at least in OpenGL) always override the driver settings. Setting the anisotropic settings in Serious Sam is nontrivial.

It would be better to see the differences in a different game that does not use application settings for anisotropic. Since the issue at hand no longer has to do with offangle surfaces (which was, thankfully, quite obviously solved by Donald's work). The issue is image quality when using the same degree of anisotropic.

 

[Dave Baumann]

Application settings for Anisotropic (at least in OpenGL) always override the driver settings.

Again - this is another case where you need to actually experience the hardware before making stupid comments (or you could even read the driver panels shown in reviews ).

ATI's drivers have an 'Application Preference' setting for Anisotropic filtering which basically means the driver Anisotropic setting is disabled, untick this and the slider is activated and it overrides any game settings.

 

[Mize]

...overrides any game settings.

Indeed, and it works even for games that would prefer it not thankfully!
Commanche 4 with 4x FSAA and 16x AF is a whole new game but only let's you turn on FSAA internally if you have a GF4 (boo).

I'm also amazed at how well the 9700 runs older games and the improvement in IQ one gets from the 9700's AA and AF. I'm actually considering installing some of the old Jane's games with their D3D patches...just might be worth it

Mize