ATi is ch**t**g in Filtering

[Chade]

:/

What do you guys make of this?

Comments?

 

[Lezmaka]

Considering the "evidence" is a bunch of screenshots taken at different locations/viewing angles, and since the poster didn't state what setting the "quality" sliders were set to, or whether trilinear optimizations for the 6800 were on/off, I wouldn't make anything of it. Get some screenshots from the same location/viewing angle, state driver version and the settings used, then a comparison can be made, otherwise, as far as I'm concerned, there are too many variables.

 

[Evildeus]

Perhaps you need to read the settings?

For all tests the Nvidia/ATI drivers were set to best image quality. This included manually setting trilinear in mipmaps and manually disabling Trilinear optimisations.

http://www.driverheaven.net/reviews/6800x800pro/testsystem.htm

 

[Chade]

:/

Considering the "evidence" is a bunch of screenshots taken at different locations/viewing angles, and since the poster didn't state what setting the "quality" sliders were set to, or whether trilinear optimizations for the 6800 were on/off, I wouldn't make anything of it. Get some screenshots from the same location/viewing angle, state driver version and the settings used, then a comparison can be made, otherwise, as far as I'm concerned, there are too many variables.

If you read later on in the thread it explains all those variables.. Surely the minor differences in the angle of the screenshots would not affect the filtering in a way that can explain the differences seen between the x800/6800 and 9800. :? If all it takes it to look at things at a certain angle to degrade the filtering that much (in comparison to the 9800) that would not be very 'intelligent' at all would it..

Hi all,

As stated in the review, drivers were set to best image quality for each test. This means that for the X800 the drivers were choosing what filtering was best. On the 6800 Trilinear optimisations were disabled, i.e. full trilinear should be on.

Regarding the positioning of the shots, those were posted to be as exact a comparison as possible during normal gameplay (i.e. i wandered round a level or two like you would when playing and took some screenshots in roughly the same areas - rather than save particular positions and try to replicate them identically using save games and the like.)...like the CMR one somethimes its V hard to get an exact comparison.

The differences in quality are however quite interesting. Maybe it would be a good idea to get someone (Icemanchilled) to create a save game file from that area, then pass it to someone with a 6800 and someone with a X800 and then compare the exact shots. Just remember that on the 6800 you must use 60.72 to see full trilinear... 61.11 dont allow disabling of optimisations. Also remember to set all other driver options and game options to High/best quality.

 

[DegustatoR]

X800 trilinear in Halo and Eclave is simply awful. It looks like bilinear actually (a driver bug maybe?).

Maybe I'll post some screenshots tomorrow, but you really have to see it in action to understand all the noises and shimmering of this "optimized trilinear".

I didn't noticed anything terrible in a whole bunch (around 20) of other games i've tested X800 on though.

 

[Lezmaka]

Thanks

I've read enough 10, 20, 30 page threads the past month or two to last me for a year. I'm just tired of reading page after page, visiting various websites just to figure out what settings are used, especially when someone is trying to make a point. If the shots had at least been from the same location/viewing angle, I might have actually looked around, but since they aren't, it seemed a complete waste of time to me. Especially after the shots from COD on xbit or digitlife or whatever the site was, seemed just like someone screwed up.

If all it takes it to look at things at a certain angle to degrade the filtering that much (in comparison to the 9800) that would not be very 'intelligent' at all would it..

Some people are saying it's not very intelligent anyway. Who knows? And that's my point. Maybe it does, maybe it doesn't, but there's no way to know for sure without using the same location/angle.

 

[Bjorn]

Some people are saying it's not very intelligent anyway. Who knows? And that's my point. Maybe it does, maybe it doesn't, but there's no way to know for sure without using the same location/angle.

This bugs me also. It's either "not exactly the same location", the screenshots are to dark to notice anything unless you change the gamma or something else that makes an immediate comparision practically useless.

 

[Lezmaka]

Wouldn't most people agree that it's pretty pointless to compare the AF of NV3x to the AF of NV4x if they aren't taken from the same location/angle? Since the method is patent pending, ATI surely won't give out any information it doesn't have to, so perhaps there's something in the algorithm that could cause a difference based on the angle. Who knows (that's willing to share anyway, heh)?

 

[Evildeus]

Sure.

 

[Tim Murray]

X800 trilinear in Halo and Eclave is simply awful. It looks like bilinear actually (a driver bug maybe?).

Maybe I'll post some screenshots tomorrow, but you really have to see it in action to understand all the noises and shimmering of this "optimized trilinear".

I didn't noticed anything terrible in a whole bunch (around 20) of other games i've tested X800 on though.

Hm, I hope somebody will try it on World of Warcraft--with NV3x, the texture shimmering is unbelievable.

 

[Hanners]

X800 trilinear in Halo and Eclave is simply awful. It looks like bilinear actually (a driver bug maybe?).

Halo has the old texture stage optimisation issue, nothing to do with 'trylinear' there. Grab a copy of rTool and set it to use trilinear on all stages and things will look much better.

 

[digitalwanderer]

rTool!?!

Dang it, I thought you said "TrayTools" last time I read you post about this and I couldn't find the option anywhere in it.

rTool!

Downloading now, thanks Hanners. I decided to give Halo another shot since I heard the patch improved it a bit but the mipmap lines were driving me a bit bonkers.

 

[chavvdarrr]

X800 trilinear in Halo and Eclave is simply awful. It looks like bilinear actually (a driver bug maybe?).

Halo has the old texture stage optimisation issue, nothing to do with 'trylinear' there. Grab a copy of rTool and set it to use trilinear on all stages and things will look much better.

Ahhhh, and obviously all users with ATi cards are smart enough to use rTool ..... it works, but why ATi don't add CP fix ???
"we believe its good enough". /sarcasm

 

[PatrickL]

Guys i can understand the interest for some of you to keep bumping a thread with that title on that forum, but don't you think that 57 pages is a bit boring ?

 

[DegustatoR]

The Baron

Hm, I hope somebody will try it on World of Warcraft--with NV3x, the texture shimmering is unbelievable.

Sorry, don't have it. Too cartoonish for me 8)

Hanners

Halo has the old texture stage optimisation issue, nothing to do with 'trylinear' there. Grab a copy of rTool and set it to use trilinear on all stages and things will look much better.

I know about this issue, Hanners. And believe me, it's not that simple. X800 gets noises and shimmering on the stage 0, the stage where all main textures are processed, not on stages >0, where all Radeons have problems with trilinear filtered bump maps.

The beginning of Silent Cartographer level have no bump maps on the ground textures, but there certainly no trilinear on X800 either.

I have 9000, 9600, 9800 and X800 Radeons. I'm getting closest to X800 quality in Halo on 9000, which can't do trilinear w/aniso at all

For now i'm tending to think, that it's a driver bug, and it will be fixed in one of the next Catalyst releases.

 

[Andrew Lauritzen]

Odd, I see nothing in those images to make me believe that the X800 is using *any* trilinear filtering, which is completely contradictory to other comparisons done so far. Possibly a game/driver bug/incompatability like we've already seen with the CoD comparison (which was using bilinear due to a bug in CoD)?

 

[Hanners]

The beginning of Silent Cartographer level have no bump maps on the ground textures, but there certainly no trilinear on X800 either.

I have 9000, 9600, 9800 and X800 Radeons. I'm getting closest to X800 quality in Halo on 9000, which can't do trilinear w/aniso at all

Well, I get the same effect on a 9800 Pro, which as you know doesn't have 'trylinear'. I use rTool to enable trilinear on all texture stages, the problem goes away. Not much more to say there.

EDIT: You can see an example of the kind of thing I'm talking about here.

 

[dan2097]

Hanners you know that tool you get from ixbt to show the differences from ATIs new filtering to trilinear, is it right that it basically makes the card think all surfaces are suitable for brilinear? If so am I correct in thinking that the surface used for testing would in the absence of this tool have actually got traditional trilinear?

 

[croc_mak]

No. The comparison is vs. uncompressed textures. I'm basically stating that it's not all that useful to use S3TC for normal map compression (since that compression assumes color data).

And no, I don't think S3TC had a tremendous impact. It is useful, and has been widely-adopted, but end users really haven't seen much difference in using them.

What?!!! 90+% of Xbox,GC and PC titles ship with compressed textures...Without which developers would have had to halve their texture resolution and the games would like crap.....Users haven't seen much difference? Can you explain what you mean here? Difference v/s what? One of the major reasons why titles of today look better than the one's released in 1998/1999 is texture resolution.....The siginificance of all other developments pale in comparision to texture resolution..period...Interestingly the most important visual improvement aspect in current and upcoming titles is the surface detail obtained by using normal maps...which also only really do the work if they are high resolution..Low-res normal maps are pretty pointless - the closer you are to 1:1 ration between your vertex count and normal map texel count, the less effective normal maps are, might as well compute lighting on per-vertex basis

My point is "surface resolution" is the most important aspect and compression techniques give you freedom to use higher resolution

 

[croc_mak]

3dc is a new compression format..Chalnoth said "it's a performance enhancement"...Isn't it logical to assume that the comparision is with the existing compression formats ie;S3TC?

No since generally ATI is marketing it against no compression. Also ati is market Parralex mapping as old normal bump mapping which is also a load of fud since parralex mapping uses hieght maps not normal maps.

Since when do we care about how ATI is marketing it? This is a technology forum..Let's talk technology..3dc is an evolution of s3tc....comparision with other compression methods is the only sensible technical discussion in my opinion.
The 3Dc format is not limited for use with "normal" maps..You can use to gain higher quality on any type of data(v/s regular S3TC). It's just a 2 component format..so, you can encode any two data elements here...