My R9700 experince and Aniso still "flawd"?

noko · Sep 9, 2002

OpenGL,

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

Donald · Sep 9, 2002

Randell said:
Donald whilst you're at it could you take those screenshots standing on one hand gargling the UK National Anthem whilst juggling a football with your feet?

Thanks

Nope....the hospital is like a 30 minute drive from my house

OpenGL guy · Sep 9, 2002

noko said:
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.

noko · Sep 9, 2002

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.

Yes, but wouldn't the triangle rotation or position determine anisotropy that a texture will have to account when filling a triangle? Which then would determine how many samples would be needed from the source texture to correctly filtered the texture for the triangle?

Really what I am asking is if the aniostropic filtering of the Radeon 9700 is based on a triangle level?

OpenGL guy · Sep 10, 2002

noko said:
Yes, but wouldn't the triangle rotation or position determine anisotropy that a texture will have to account when filling a triangle? Which then would determine how many samples would be needed from the source texture to correctly filtered the texture for the triangle?

All you care about is du/dx, dv/dx, dv/dy and dv/dy right? The triangle rotation or position doesn't matter, what matters is how the texture is aligned with the pixels.

Really what I am asking is if the aniostropic filtering of the Radeon 9700 is based on a triangle level?

I don't think so.

Nagorak · Sep 10, 2002

jb said:
Geez

I had a funny feeling that this would happen. Sorry Dave I was only going to PM this to you to see if you could validate what I saw. Did not mean to turn it into a hot topic. I just thought that maybe a few others that I know have a 9700 could help. But so far the 9700 has been a champ and thought that was worth sharing. My bad Most site said that the aniso issue was fixed. So when I saw it in SS:SE I was confused.

Folks I ran that level a few minutes and did saw that "biggest" diff was around 45 degrees. There was not a "big" change till then.

However I have ran it though all the other FPS shooters I have and not once did I see any other issues with Aniso. This is a very nice card and I am happy with it.

Could it just be a bug with the game then? I know some games don't like it when you try to force AF (Deus Ex, for example, the floor keeps flashing between AF and blurry). Maybe there's something going on either in the game engine or in the drivers.

KimB · Sep 10, 2002

OpenGL guy said:
noko said:

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

Click to expand...

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.

I'd like to tackle this with a little supposition, based upon the following:

1. The MIP map level chosen can be dependent upon the degree of anisotropy.
2. The MIP map lines on the 8500 don't appear to be simply function-based (i.e. a result of precision that is too low), but look, to me, to be more of a triangle-based approach. I really can't see how else you'd get "sides of a polygon" as the MIP map boundaries. I really do not understand how a MIP line like that could occur on a pixel-based algorithm.

Because the MIP map selection algorithm looks more like a poly-based method (as, if I remember correctly, nVidia called the TNT's method), and since MIP map selection apparently depends upon the degree of anisotropy, it only makes sense that anisotropy would come very near MIP map selection, meaning it just makes sense for it to be a per-poly approach (likely done within the triangle setup engine).

OpenGL guy · Sep 10, 2002

Chalnoth said:
A bunch of rubbish.

Chalnoth,

Why don't you quit while you're ahead? You can't figure something out (I really can't see), so why even bother to post? You're completely wrong by the way. I'll pull out my favorite link to prove it:
http://www.opengl.org/developers/do...ec1.1/node83.html#SECTION00681000000000000000

Take note of the comment about approximations immediately following equation 3.12.

Now, I ask you to plot the graph of y = |x|. What does it look like? What do you know! It looks VERY similar to the R200's miplines! Coincidently, it looks like the sides of a polygon, but it's just a coincidence.

If people aren't going to learn from the information posted to this board, that's fine: You can take a horse to water, but you can't force it to drink. However, to post such blatant, incorrect information is just plain wrong.

tEd · Sep 10, 2002

Bigus Dickus said:
A couple of comments... since so many others have addressed Chalnoths post, I see no need to do so. I was shocked when I read it, for obvious reasons. Second... just what were the settings in the GF4 shot? Either the LOD is reduced, or Quincunx was used, becasue it really does seem to be a "blurry mess" as someone else described it. I'm assuming this is due to some particular setting used, so I'm interested in knowing which was responsible.

neither LOD was increased nor was quincunx used in those gf4 shots. There is a filter option in ss where you can set a spezial filter to sharpen the textures this option was set to normal mean no additional filter was applied. The simple reason why I choose this is that the sharpen filter for me add texture aliasing on some textures what I don't like much

jb · Sep 10, 2002

Nagorak,

thats what I thought at first too. But when I saw the same thing with that D3D program that pccen wrote (I probably did not spell his name right) then I was confused....

Bigus Dickus · Sep 10, 2002

tEd said:
neither LOD was increased nor was quincunx used in those gf4 shots. There is a filter option in ss where you can set a spezial filter to sharpen the textures this option was set to normal mean no additional filter was applied. The simple reason why I choose this is that the sharpen filter for me add texture aliasing on some textures what I don't like much

Hmm... just a higher default LOD then I guess?

BRiT · Sep 10, 2002

OpenGL guy said:
If people aren't going to learn from the information posted to this board, that's fine: You can take a horse to water, but you can't force it to drink.

No, but you could hold their head under water. That way you wont have to deal with them anymore...

--|BRiT|

OpenGL guy · Sep 10, 2002

BRiT said:
No, but you could hold their head under water. That way you wont have to deal with them anymore...

That might save my sanity, but I don't want my next valentine to be some guy named Bubba

noko · Sep 10, 2002

The triangle rotation or position doesn't matter, what matters is how the texture is aligned with the pixels.

I can see that, at least I think I do. Still, doesn't the position of the triangle to be filled determine the pixel layout for the texture? So are you saying by looking at the final pixel layout for the on screen texture you can determine what anisotropy would be required for filtering?

In other words lets say I have a polygon in which the back is 10 pixels across while the closer side of the polygon is 80 pixels across. For a texture to fit on this it has to be squashed by a factor of 8x from front to back, is that correct?

So using a simpler calculation I can say this 8x factor is a anisotropy of 8x, in this case I will need at least 8 filtered textures from the source texture to correctly render or map this polygon in the frame buffer or on the screen. Do I have this correct or am I way off?

OpenGL guy · Sep 10, 2002

noko said:
I can see that, at least I think I do. Still, doesn't the position of the triangle to be filled determine the pixel layout for the texture? So are you saying by looking at the final pixel layout for the on screen texture you can determine what anisotropy would be required for filtering?

As I said before, what's important is how the texture is matching up to the pixels. Think about du/dx, dv/dx, du/dy, and dv/dy. These only relate the texture coordinates (u,v) to the pixel coordinates (x,y).

In other words lets say I have a polygon in which the back is 10 pixels across while the closer side of the polygon is 80 pixels across. For a texture to fit on this it has to be squashed by a factor of 8x from front to back, is that correct?

The rate of change is important, too. Thus, how long are the sides? For example, if the bottom were 80, the top 10, but the sides were 1000000, I think you'd agree that the degree of anisotropy is rather small.

So using a simpler calculation I can say this 8x factor is a anisotropy of 8x, in this case I will need at least 8 filtered textures from the source texture to correctly render or map this polygon in the frame buffer or on the screen. Do I have this correct or am I way off?

I don't want to get into a detailed discussion of anisotropic filtering techniques (there are better sources for the information than what can be covered here). See:
http://www.meko.co.uk/anisofilter.html
http://home.swipnet.se/~w-12597/3dxtc/articles/anisotropic.htm
for more information. (These are just a couple of links I found in Google, there may be better ones around.)

tEd · Sep 10, 2002

Bigus Dickus said:
tEd said:

neither LOD was increased nor was quincunx used in those gf4 shots. There is a filter option in ss where you can set a spezial filter to sharpen the textures this option was set to normal mean no additional filter was applied. The simple reason why I choose this is that the sharpen filter for me add texture aliasing on some textures what I don't like much

Click to expand...

Hmm... just a higher default LOD then I guess?

no it's not that.

just the sharpen filter in serious sam.

that's how it looks on a gf4 with sharpen filter at max

http://homepage.sunrise.ch/homepage/rstude/ss2gf4_4.jpg
http://homepage.sunrise.ch/homepage/rstude/ss2gf4_5.jpg

Donald · Sep 10, 2002

Ok kids here we go.

Game settings:

Serious Sam: SE
1024x768 32-bit
Quality settings selection (nothing else touched)

Systems:

Main workstation:
P4 2.0AGHz
Asus P4s533 SIS 645DX
9700 Pro
7.75 Catalyst drivers (6143 off CD)
Tweaks done via ATI control panels

Server:
P3 1.13GHz
Abit VH6T VIA Apollo Pro (/rant on...curse you Intel for not supporting more than 512MB and making me buy VIA shite!.../rant off)
Gainward Ti4200 (normally a GF4 MX)
30.82 Detonator drivers
Tweaks done by Rivatuner

All shots were taken with quality anisotripic selected for each card and 4x/8x used for 4200 with 8x/16x used for 9700. Shots also reflect the default LOD values for respective drivers. The labeling is reflective of the angle I took the shot at. 45 is 45 degrees, 22 is approximately 22 degrees and level is...well level. All of these shots should be REALLY close. Croteam was nice enough to have a very straight line in the texture to use as a reference point for the 22 degree shot

I sure hope I didn't make any typos anywhere

Oh and does anyone know of a way to capture a game you are playing to video so I could post up how things look in motion?

Edited for typos!

OpenGL guy · Sep 10, 2002

Donald said:
Ok kids here we go.

Thanks, Donald.

I did get one broken link:
http://www.thoroughbred-data.com/nudies/ssaniso/ss970042008x1.jpg

Donald · Sep 10, 2002

OpenGL guy said:
Donald said:

Ok kids here we go.

Click to expand...

Thanks, Donald.

I did get one broken link:
http://www.thoroughbred-data.com/nudies/ssaniso/ss970042008x1.jpg

Most welcome and that link is now fixed, thanks for catching it.

Galilee · Sep 10, 2002

Great work Donald.
Looks like the image on R300 change a little, but becomes similar to GF4 in it's worse position (that looks like 22 degrees to me).
When both cards have the non-rotated floor they look the same except faar away where R300 is sharp up to the wall, while GF4 becomes blurry a few meters from the wall.

My R9700 experince and Aniso still "flawd"?

noko

Donald

OpenGL guy

noko

OpenGL guy

Nagorak

KimB

OpenGL guy

tEd

Casual Member

jb

Bigus Dickus

BRiT

(>• •)>⌐■-■ (⌐■-■)

OpenGL guy

noko

OpenGL guy

tEd

Casual Member

Donald

OpenGL guy

Donald

Galilee

Similar threads