"Trilinear" Filtering in Motion

[Bouncing Zabaglione Bros.]

DT, I think you're stretching things a bit far. Do not attribute to conspiracy that which can be explained by ignorance and stupidity.

That would be a valid viewpoint if if wasn't for the Nvidia PR slides plugging the *same* speed increase in the *same* application detected benchmark. That blows the "accidental bug" theory out of the window. It's a deliberate cheat, just like the others we've seen.

 

[BRiT]

Oh, BZB, I'm not using that to describe Nvidia themselves. Everything they do is for a specific reason -- to try to regain or limit the lost market/mind share. But it (ignorance and stupidity) does seem to fit the situation when they (K/[H]) claim "I can't see the difference." They haven't a clue as to what they should be looking for.

I would go ballistic if I spent my money on a card that produced the crawling and obvious visual anomolies. Every instance and situation that occurs makes me glad I spent the $230 USD 10 months ago to pick up a 9700Pro instead of waiting to see what Nvidia had in the wings.

 

[StealthHawk]

Yes that is true, or even long stairs like "Halls of Anubis", I've already saw the exact same thing when I played around with a friends 5600.

Static screen shots do not show the truth, like [H] tried to mislead its readers with.

But...but...but....[H] said that the mip maps weren't visible in motion either!

edit: [H] thread = People are saying the comparison is invalid because there is too much compression on the videos If anything, that would mask the bilinear artifacts :?

 

[digitalwanderer]

Yes that is true, or even long stairs like "Halls of Anubis", I've already saw the exact same thing when I played around with a friends 5600.

Static screen shots do not show the truth, like [H] tried to mislead its readers with.

But...but...but....[H] said that the mip maps weren't visible in motion either!

edit: [H] thread = People are saying the comparison is invalid because there is too much compression on the videos If anything, that would mask the bilinear artifacts :?

ROFLMFAOTTRIME~~~~~~

 

[MikeC]

A few more video clips from the NV35.

On a bridge with limited lighting:
http://www.nvnews.net/temp/1-1.avi - 3.5 MB

On the sand with full sunlight:
http://www.nvnews.net/temp/1-2.avi - 3.1 MB

Original video shot at default movement speed:
http://www.nvnews.net/temp/1-normal-speed.avi - 2 MB

 

[BRiT]

Getting not found for the 1-1 and 1-2 DiVX...

And even at full speed, i cant stand the way the detail rolls in. *shiver*

 

[MikeC]

edit: [H] thread = People are saying the comparison is invalid because there is too much compression on the videos If anything, that would mask the bilinear artifacts :?

Yes, I read the comment. I compressed the files in DivX format, which was much worse in quality compared to Microsoft's MPEG4. It's possbile that they aren't playing the video back at it's full resolution of 800x600.

Anyway, I have a couple of uncompressed videos that I'm moving to the server right now. Each file is around 25MB, which is a couple seconds of video but just enough to see the differences. You'll be surprised that the quality isn't significantly better than the MPEG4 files.

I'll update this post when the uncompressed videos have been moved to the server.


http://www.nvnews.net/temp/1-no-compression.avi - 25.9 MB

http://www.nvnews.net/temp/2-no-compression.avi - 25.2 MB


Probably a good idea to right click on the links and choose "Save As..."

 

[MikeC]

Getting not found for the 1-1 and 1-2 DiVX...

Guess it would help if I moved them to the server Give me about 10 minutes since I'm moving the uncompressed videos over now.

 

[digitalwanderer]

Hmmmm....me wonders if MikeC's servers monthly bandwidth usage will be a tad higher this month....

Thanks Mike, d-loading it now.

 

[reever]

Its really annoying to have these people complaining because the image is more compressed. Even if it is of low quality, grainy, compressed and you STILL can see the difference, a higer quality video would only make the difference more clear? right?

 

[Pete]

Mike, you were The Man enough for your extra-effort FRAPS benchmarking, but sharing your bendwidth-intensive findings in a forum not your own deserves more props than I know how to give. Cheers (even if you may be sharing in B3D because it has a smaller audience, and thus smaller bandwidth bill )!

PaulS, I figured the sloping/curved ground may make the MIP-maps more pronounced, just wanted to be sure. Thanks for the supposed confirmation.

Edit: I can't see the MIP-map lines in the latest videos.

 

[digitalwanderer]

MikeC: What codec do I need for the uncompressed ones? Media player ain't having any luck with 'em.

 

[nate240]

Well, try as I may, I cant see the mip lines in the full speed vid, of course the grass and length of the vid make it harder. I can, however, still clearly see the mip lines in the bridge vid, and can barely make them out in the beach vid.

 

[Ostsol]

MikeC: What codec do I need for the uncompressed ones? Media player ain't having any luck with 'em.

WindowsXP reports the video codec as Fraps. Do you have FRAPS 2.0 downloaded and installed?

Well, try as I may, I cant see the mip lines in the full speed vid, of course the grass and length of the vid make it harder. I can, however, still clearly see the mip lines in the bridge vid, and can barely make them out in the beach vid.

Watching the first video I could see that mip-mapping was occuring, of course, and noted a point that I thought was a mip-map border. Watching the second video I noted two points that I had missed entirely in the first video closer to the POV that were clearly mip-map borders. Going back again to the first video I finally realized that there were other borders I had not noticed before. I find that the smooth transition -really- helps.

 

[Brent]

Can I ask a question?

Do you play the game with No AF?

Maybe some comparison videos with AF enabled would be more relevant to what settings people would use on mainstream - high end cards, no?

 

[Ostsol]

That's the same rationalization ATI used when question regarding the lack of trilinear filtering in conjunction with AF on the Radeon 8500. Ironic that NVidia must use it now. . .

In any case, there are sufficient instances where even maximum AF does not push the the first mip-map border out of view that the matter of trilinear versus bilinear can be significant. I noticed this in Max Payne and UT on my Radeon 8500. Mip-map borders were visible in enough instances at 8x AF that I considered using trilinear filtering with no AF. Of course, in that case I would easily note the individual mip-map levels, even though the borders were blurred.

The matter of whether it is easily visible or not is not what bothers me most, anyway. It is a matter of controlling graphical settings. Do you believe it is right for NVidia to take away the option to use the application's settings for texture filtering? Also, is this so different from Quack?

 

[Archaeolept]

that's fair enough. Higher AF/resolution will push this "rolling wave of distortion" bilinear boundary further back such that it shouldn't really be noticable - exactly like what happened w/ ATI's R200 core, where the lack of trilinear along w/ AF wasn't really that great a loss.

but, does this mean then that you won't be benching at non "high-quality" (low AF, and maybe no AA) settings anymore? just kind of following the implication as to what is important when reviewing cards...


edit: damn ostsol beats me to it

 

[madshi]

Hi Mike,

have you seen this?

http://www.beyond3d.com/forum/viewtopic.php?p=145076#145076

Dave posted some screenshots there, where mipmap level seem to clearly show. Perhaps you can try to check out the locations where he shot those screens for another video? Thanks!

 

[StealthHawk]

Can I ask a question?

Do you play the game with No AF?

Maybe some comparison videos with AF enabled would be more relevant to what settings people would use on mainstream - high end cards, no?

Brent, read Ostol's post below yours.

Also, I would like to chime in here. In my main rig I have a r9500Pro. I have used both Performance AF and Quality AF. As you know, Performance AF uses bilinear only, and Quality AF uses trilinear in stage 0, bilinear in the other stages, which is fine for most games.

I can tell the difference between Performance Af and Quality AF. I see absolutely ZERO difference in screenshots. But in motion, I can see the mipmap transitions with 16x Performance AF, where I see none in 16x Quality AF. I hope this answers your question.

I thought the whole UT2003 issue was quite clear from the beginning. NVIDIA's shortcuts are apparent in some situations, and in others, they are not. Clearly speaking, yes, you can find places where it probably will not be noticeable at all. But you can also find places where it is noticeable. The fact that some people may not be able to see the difference is pretty irrelevant. If some people CAN see the difference, and want the higher quality, the forcing of lower quality is absolutely indefensible.

edit: It also seems strange to me that now you are claiming that there is no visible difference between bilinear AF and trilinear AF. Especially when you said this about Performance vs Quality AF in UT2003:

The Radeon 9700 Pro and Radeon 8500 at 16X anisotropic are very smooth. I really don’t see much difference here in anisotropic quality between the Radeon 8500 and Radeon 9700 Pro, though. Do keep in mind though that the Radeon 9700 Pro is using Trilinear filtering with anisotropic while the Radeon 8500 is only using Bilinear. In game the difference is noticeable as you move if you look for it.

http://www.hardocp.com/article.html?art=MzQ0LDQ=

And yet are you now implying that there is no difference between bilinear and trilinear when only 8x AF is used?

It also seems strange that in a previous article people at [H] could tell the difference between full trilinear and tri/bi. Albeit you weren't the author of that article. http://www.hardocp.com/article.html?art=NDQ0LDE=

Interestingly enough, you gave the nod to the gfFX when no AF was applied in your UT2003 article. Are you saying you cannot see the difference between video 1 and video 2?

In-Game Still Screenshot No AF = 5900 Ultra SLIGHTLY Sharper Textures with no mipmap transitions that are distracting.

In-Game Movement No AF = Can’t tell any difference.

http://www.hardocp.com/article.html?art=NTAwLDQ=


And of course, I am still wondering whether or not trilinear AF was really applied for the r9800 since you configured UT2003 this way:

The ATI driver control panel was set to “Application Preferenceâ€￾ on both AA and AF so that we could test with AF disabled. Then we set the AF slider to Quality AF when AF was tested.

 

[Brent]

In any case, there are sufficient instances where even maximum AF does not push the the first mip-map border out of view that the matter of trilinear versus bilinear can be significant.

Oh

I guess I just haven't come across, or noticed yet a transisition with the first mip-map border with 8XAF enabled.

Do you believe it is right for NVidia to take away the option to use the application's settings for texture filtering?

Who am I to decide if what NVIDIA is doing is wrong or right, it really isn't my place to decide wrong or rightness. According to what I hear developers say they would really like it that the drivers not change up things like that, and render how they intended the game to be. That is also what the end user is asking for. So it seems clear, most people would like it to use application filtering as default, and then maybe have an option for faster performance with the expensive of lowering iq a bit.

I really keep questioning myself with this, wondering why the heck we are still having to look at something like Trilinear filtering on these high-end cards of today (the Trilinear filtering method has been around for a long long time now). We should be concentrating solely on Anisotropic methods and higher levels of it with today's cards, but here we are still looking at an old filtering method, we just seem to not be able to get past it

Also, is this so different from Quack?

IMO it does sound similar. But I stayed out of the whole quak issue when it went down, so i don't really have an excactly clear picture all the details from all camps. I got the impression ATI said it was a bug, and when they fixed it performance didn't drop, it stayed the same. So if performance stayed the same with better IQ, then thats an optimization. So maybe it was an optimization gone bad, that had an un-intended side-effect at first, once they fixed it, iq was better, speed stayed the same.

Thats the impression I get when I read about it now.