Bilinear and trilinear

K.I.L.E.R

Retarded moron
Veteran
I tried playing with bilinear and 16x AF on my R300 and I noticed mipmaps pretty damn well, mipmaps aren't the worst part.

When you move around you will notice parts become blured while others become sharp and sometimes too sharp causing texture aliasing?

On my brothers NV20 it is far more noticeable. Screenshots don't show that part because you have to be moving around to notice it. :LOL:

On my R300 it's far less noticeable compared to the NV20, that would probably be because I'm using 16x AF and a negative LOD.

With quality selected this doesn't seems to be a problem anymore.

So how do people (especially some review sites) come to a conclusion that there isn't a noticeable difference between trilinear and bilinear?
Even 1 stage trilinear makes a very large difference on both pieces of hardware.

Is it because others are partially colourblind? Sorry for bringing colour blindness into this but there is simply no other explanation that I can think of.

I have seen screens here at Beyond3D as well as [H] and other sites and people claiming ones better than the other but it looks to me that everyone is either trying to show the worst case scenario on screenshots or best case scenario.

It's so easy to do that when standing still. I paused Postal 2 in the bank area and the picture looked perfect while paused but when I move around, oh that's a completely different case alltogether. You start to notice some places become blured while others too sharp and some just right.

If you increase the LOD bias you will find that the blur becomes far more noticeable in some areas.
Of course it's still better than playing with no AF, nonetheless bilinear looks horrible regardless of which way you view it.

I thouht the entire point of buying a high end card was to play with max IQ and no compromises (unless the user wants to create some)?
 
I dont know if anyone has tried this yet, but how about utilizing .gif images to show the different levels of filtering in action. Or better yet, attempt to make a brief, high-quality video that demonstrated such a thing? I know that GIFs can be quite large, especially if you create one running at least at 30 FPS for a brief period of time at a resolution of 800x600+.

I know this is asking a bit much of reviewers, but it would put to rest all of these problems arising from subjectively reviewing AF quality or simply posting pictures, as seeing mip-map boundaries in motion could give readers more information.

It sounds like a good idea, on paper that is. ;)
 
Yeah, I really dislike AF on my R8500 on most games. Especially when watching Colin McRae Rally 3 replays, the texture aliasing is horrible. If the camera is still, it's good, but when it starts to move... Ugh..
 
Lack of trilinear is not the only cause of R200's texture aliasing.
There are artifacts like mipmap boundaries even where the miplevel doesn't change (using color mipmaps can easily reveal this)

The card uses insufficient samples for AF and so you can see where it swiches between the number of samples used.

R300 doesn't do this (well, at least in D3D ...), so it has much better quality with bilinear AF than the R200.
 
Nonetheless bilinear filtering is very poor compared to trilinear filtering. Why can't review sites mention the artifacts in bilinear which trilinear filtering doesn't have?
 
PaulS said:
I don't think colourblindness has anything to do with it.

And i should know :?

There are far too many levels of colourblindness, so you aren't able to generalise. Your CB condition would be different from another perons.

Some people have trouble detecting some shades while others have trouble with different set of shades.

If all you can see is black and white maybe most artifacts of blinear filtering would be gone? ;)
 
Squega said:
I dont know if anyone has tried this yet, but how about utilizing .gif images to show the different levels of filtering in action. Or better yet, attempt to make a brief, high-quality video that demonstrated such a thing? I know that GIFs can be quite large, especially if you create one running at least at 30 FPS for a brief period of time at a resolution of 800x600+.

I know this is asking a bit much of reviewers, but it would put to rest all of these problems arising from subjectively reviewing AF quality or simply posting pictures, as seeing mip-map boundaries in motion could give readers more information.

It sounds like a good idea, on paper that is. ;)

GIFs only allow 8-bit color. (Actually, they support 24-bit color, but only without compression, and I'm not sure you can have animated true-color GIFs in theory and in practice they load too slowly to be animated.) So, no good.

MNGs (the animated form of PNGs) might be a good solution, although browser support is spotty. And I wonder if full screen true color playback at anywhere near in-game framerates is at all plausible with current hardware.

Divx (or some such) at high bitrates may be an option, but I would think that lossy compression, even at a high bitrate, is going to make evaluating texture aliasing difficult.

The other problem is actually capturing lossless video feeds of a game in motion. 1600x1200 @ 100fps works out to 549 MB/sec. Sorry, but you're not going to be able to write that to your hard drive at that pace. Now, 800x600 @ 30fps is a barely managable 41 MB/sec. And indeed I think FRAPS now supports video capture up to that limit.

So, it may be possible, but just barely.
 
K.I.L.E.R said:
Nonetheless bilinear filtering is very poor compared to trilinear filtering. Why can't review sites mention the artifacts in bilinear which trilinear filtering doesn't have?

:?:

I don't think any review sites benchmark in bilinear at all anymore. The recent controversy is that with 44.03 drivers, NV3x is using what appears to be the old "balanced" settings when running UT2003 at "quality" settings. This means proper LOD and mixed bi/trilinear (blending mipmaps only near mipmap transitions).

The issue may or may not be noticeable in-game. [H] says they tried to look for it and couldn't tell. Dave says he can tell. Some screenshots show it (subtly), and some don't.

Probably the issue is noticeable in many in-game environments, particularly if one is looking for it. But I think the one thing everyone agrees on is that it is nowhere near as noticeable as straight bilinear.

And while [H] didn't mention the issue in their recent 5900U review (after it had been brought to Brent's attention), they later dedicated an entire article to the optimization's effects on image quality. Just because their conclusion was that those effects weren't noticeable doesn't mean they ignored the issue.
 
I use the utility rTool to force tri-linear sampling on all mipmaps when forcing AF. I noticed in Mafia even with 16x quality for AF that mipmap lines was present indicating the more distant mipmaps where indeed bilinear filtered. With rTool you can have all the mipmaps on a R300 plus card Tri-lineared if you want. I really didn't notice that much of a performance penalty by forcing Tri-linear but I did notice a big IQ improvement (no moving mipmap lines!). You can download from here:

http://exxtreme78.bei.t-online.de/DE/rTool-setup.exe

rtool.gif
 
You really don't need this tool to force Trilinear in all the mipmaps when forcing AF, but it is rather convient to use rTool vice editing. In the registry you can set the filtering method used for forced AF by the following: (Just do a search in RegEdit for AnisoType for DirectX or OGLAnisoType for OpenGL:

DirectX
  • AnisoType = 0 (This will give all Tri-linear filtered samples for AF)
    AnisoType = 1 (Performance Setting. This will give all Bi-linear filtered samples for AF)
    AnisoType = 2 (Quality Setting. This will give a combination of Tri and Bi samples for AF)

OpenGL
  • OGLAnisoType = 0x00000000 (0) (All Tri-lineared samples)
    OGLAnisoType = 0x00000001 (1) (All Bi-lineared samples)
    OGLAnisoType = 0x00000002 (2) (Combination of Tri and Bi)

The drivers do exposed the third higher quality AF setting and I hope ATI will keep this ability and exposed it in the control panel in the future. It is well worth the IQ increase. Maybe ATI can call it SuperQuality or something :).
 
noko said:
I use the utility rTool to force tri-linear sampling on all mipmaps when forcing AF. I noticed in Mafia even with 16x quality for AF that mipmap lines was present indicating the more distant mipmaps where indeed bilinear filtered. With rTool you can have all the mipmaps on a R300 plus card Tri-lineared if you want. I really didn't notice that much of a performance penalty by forcing Tri-linear but I did notice a big IQ improvement (no moving mipmap lines!). You can download from here:

http://exxtreme78.bei.t-online.de/DE/rTool-setup.exe

http://bellsouthpwp.net/n/o/noko/rtool.gif
The rTool cannot FORCE full Tri-AF. When an application wants Bi-AF so there will be Bi-AF when you switch to "trilinear AF" in the old rTool-Version. :)

Unfortunately the ATi-driver itself cannot force full Tri-AF.

And if you want to choose between the TS-optimized Tri-AF and this "Application"-Tri-AF so get the newest version here:
http://exxtreme78.bei.t-online.de/DE/rTool-setup_0.9.9.7b.exe
 
COOL Thanks for the update. Well in Mafia your older version did indeed force Tri-Linear. The mipmap lines was very apparent with the drivers force AF even in quality mode. Now rTool did a nice job in making them dissappear with a small drop in FPS. I will try out your newer versions and see if I can tell any other difference. Thanks again.
 
Back
Top