My R9700 experince and Aniso still "flawd"?

[Mize]

The 9700 is a huge IQ improvement over the 8500.

As I once posted on Rage 3D: There is no time when the 9700 doesn't apply anisotropic filtering. Unless, of course, no aniso is needed.

I'm not saying there are points where aniso isn't applied, simply that the mipmap boundaries move closer to the pov at regular angular intervals. If you don't have a copy of pcchen's mipmap program send me your email address and I'll send it to you so you can see what I'm talking about. I'll also fire up SS:SE later tonight to look at that rotating floor scene.

Mize

 

[OpenGL guy]

I'm not saying there are points where aniso isn't applied, simply that the mipmap boundaries move closer to the pov at regular angular intervals. If you don't have a copy of pcchen's mipmap program send me your email address and I'll send it to you so you can see what I'm talking about. I'll also fire up SS:SE later tonight to look at that rotating floor scene.

I wasn't arguing with you: I know exactly how it's supposed to look. I was just pointing out that this is not new information.

 

[KimB]

There are minima in the level of aniso filtering starting at about 15 degrees and every 45 thereafter. It is nothing like the minima in anisotropic filtering seen on the 8500; in this case the first minima is at about 45 degrees and repeats every 90 degrees and is quite severe (vitually no filtering). Personally, I find 16x + trilinear on the 9700 to be about the best IQ I've seen and at an amazingly low hit.

Mize

Very interesting. In particular, this seems very analagous to the improvement in the MIP map lines seen in the 9700 (They didn't move to a smooth function...just increased the number of segments on the polygon used for the boundary...).

I just do not understand why ATI can't seem to implement smooth functions, instead of the "edges of a polygon" type approaches they seem to have been implementing. After all, nVidia's had 'em since the original GeForce...why hasn't ATI done it?

 

[Bigus Dickus]

I just do not understand why ATI can't seem to implement smooth functions, instead of the "edges of a polygon" type approaches they seem to have been implementing. After all, nVidia's had 'em since the original GeForce...why hasn't ATI done it?

Does it offer a visual improvement (other than prettier looking colored mip-map screenshots)? Does it result in any loss of performance?

 

[OpenGL guy]

I just do not understand why ATI can't seem to implement smooth functions, instead of the "edges of a polygon" type approaches they seem to have been implementing.

Can you even tell the difference, without using colored mipmaps? Oh wait, you don't own an ATi card.

After all, nVidia's had 'em since the original GeForce...why hasn't ATI done it?

Maybe the designers think there are better ways to spend gates.

 

[KimB]

Does it offer a visual improvement (other than prettier looking colored mip-map screenshots)? Does it result in any loss of performance?

Well, without anisotropic filtering enabled, I do know that I noticed a huge difference when bilinear filtering between my GeForce card and my TNT (the TNT used a MIP line similar to the Radeon 8500's...).

As for the tangible difference the MIP map lines make, particularly with the improved lines on the 9700's, I doubt it will be visible. Regardless, this wasn't the reason I brought up the MIP lines. I brought up the MIP lines because it appears that the correlation between how the MIP boundaries changed and the anisotropic off-angle problem changed is just too similar to be a coincidence. As another point of comparison, the GeForce series has no off-angle problem, and also has smoothly-curved MIP lines. It seems likely to me, therefore, that when ATI finally goes for smoothly-curved MIP lines, the aniso problem will be solved.

At the same time, the off-angle anisotropic problem on the Radeon 9700 will still be visible in certain specific scenarios. Those in Serious Sam: SE were already pointed out, but there's also the obvious flight sim issues.

With over 100 million transistors, why couldn't ATI finally actually solve the problem, not improve on their approximation?

 

[Althornin]

Give it a rest chalnoth.

It seems likely to me, therefore, that when ATI finally goes for smoothly-curved MIP lines, the aniso problem will be solved.

Sorry, but quite frankly, YOU DONT KNOW WHAT YOU ARE TALKING ABOUT.
You are basing all of this on secondhand data, from OINE data point.
Get real.
As for their aniso algorithm - isnt it obvious?
They want to offer HIGHER PERFORMANCE with almost no visual degredation.
You, on the other hand, only want to make ATI look bad.
Look at those spiral number pics.
Can you see any aniso inconsistency?
Oh wait, that doesnt jibe with your current "ati sucks theory of the week".

 

[alexsok]

Sorry, but quite frankly, YOU DONT KNOW WHAT YOU ARE TALKING ABOUT.
You are basing all of this on secondhand data, from OINE data point.
Get real.
As for their aniso algorithm - isnt it obvious?
They want to offer HIGHER PERFORMANCE with almost no visual degredation.
You, on the other hand, only want to make ATI look bad.
Look at those spiral number pics.
Can you see any aniso inconsistency?
Oh wait, that doesnt jibe with your current "ati sucks theory of the week".

He wants to make ATI look bad?

Why don't u reread his posts, u quite frankly misunderstood him.

 

[Althornin]

He wants to make ATI look bad?

Why don't u reread his posts, u quite frankly misunderstood him.

No, i dont think i did.

 

[Nagorak]

I looked at the pics above and quite honestly didn't see anything wrong with them. Please don't try to point it out...I'm not really THAT interested in having to look at a pic zoomed in 10x to see that one pixel is slightly incorrect...

It seems that people are really grasping at straws with this IQ stuff. Please, if you don't like your R9700 I'll take it off your hands for a cool $200!

 

[PCGH_Thilo]

The 9700 is a huge IQ improvement over the 8500.

As I once posted on Rage 3D: There is no time when the 9700 doesn't apply anisotropic filtering. Unless, of course, no aniso is needed.

I'm not saying there are points where aniso isn't applied, simply that the mipmap boundaries move closer to the pov at regular angular intervals. If you don't have a copy of pcchen's mipmap program send me your email address and I'll send it to you so you can see what I'm talking about. I'll also fire up SS:SE later tonight to look at that rotating floor scene.

Mize

mize, is the rotating floor scene in the technology level?

 

[Gery]

Well, without anisotropic filtering enabled, I do know that I noticed a huge difference when bilinear filtering between my GeForce card and my TNT (the TNT used a MIP line similar to the Radeon 8500's...).

Sorry, but that is wrong. The TNT uses only a per polygon mipmapping, so the Mip-Line was identical to the polygon edges. Additional it with trilinear filtering enabled it uses an very ugly dithering algorithm, compared to the dithering from other cards like the 3dfx-Voodoo1.

The 8500 uses per pixel mip-mapping, but the mip-line is calculated very rougly (it seems it has to be a maximum of around 20° off from the polygon edges).

 

[tEd]

@pcgw_thilo

no it's in the level "the pit" or ,"das loch"

If anybody wanna try out , i have a save game for that particular point here

http://homepage.sunrise.ch/homepage/rstude/ss2save.zip

it's from the european version , i hope the savegame also works with us vers.

 

[darkblu]

Why wouldn't this same issue be seen on the skewed walls on the non-rotated shot?

Aren't those triangles at an odd angle, or is there something I'm missing?

Russ,
essentially, the exhibited level of anisotropy on those skewed wall is considerably lower than that of the floor. plus the big luminocity variance across those walls hampers the observation.

 

[Donald]

Hi everyone.

jb contacted me thru 3dgpu (I have a 9700) and asked me to verify his findings on the 45 degree aniso problems. Here are some shots that I just finished taking at the same place in SS at 1024. I used quality settings in the game, made sure trilinear was enabled in game as well. I also used the 16x quality anisotropic filtering option along with 4x AA in the ATI control panel. Due to size (about 250K) and consideration for dialup users I'm just going to provide direct links. These are in OGL btw:

http://www.thoroughbred-data.com/nudies/ssaniso1.jpg

http://www.thoroughbred-data.com/nudies/ssaniso2.jpg

http://www.thoroughbred-data.com/nudies/ssaniso3.jpg

Basically it's still there

 

[jb]

Donald,

thanks sir. But if you look on page 1 of this thread you can see the 8500 shoot which is much worse.

 

[OpenGL guy]

Basically it's still there

What's still there? Did you compare your results to plain-old bilinear?

People are making these comments without comparing them to the baseline (i.e. bilinear).

 

[Donald]

Yes, it is MUCH improved over the 8500, I guess I should have noted that eh?

 

[Donald]

Believe me OpenGL Guy, I didn't mean to offend. I just forgot to add something to the end of the post, nothing more.

 

[Mize]

The rediculous thing about this thread is that the closer mipmap shots from SS:SE look like 8X AF on a GF4 and yet people are whining.

Mize