AF again

[Dave Baumann]

This is what bit me when doing this piece. The AF setting is buried in the bowels of the Advanced Rendering settings, and it supercedes whatever driver setting you're trying to force on (on nVidia's 40.41 driver anyway).

FYI, probably the best thing to do to make sure of consistent output in SS:SE is to select the 'Extreme' Settings in the 'Addons' and then select the level of Aniso desired in the 'Advanced Rendering Options' -- again, remembering to turn Aniso to 'Application Select' on Radeon's.

 

[noko]

The article states:

ATI's implementation of AF in the Radeon 9700 takes the slope information calculated during triangle setup (scan-line conversion) and retains it for use later in the pipeline for texture filtering operations. This slope information is determined on a per-polygon basis. ATI's approach not only determines whether to use AF or not, but also how much AF to apply, depending on how steeply sloped a surface is relative to the view camera.

This is incorrect, this was discussed before on this forum and ATI does it per pixel not per polygon. The AF requirements will vary in a polygon. Now Nvidia may do per polygon but ATI doesn't. This could very well be the reason why there is less performance loss with the ATI method over Nvidia's.

Updated Final Thoughts:

Our conclusion: compared to filtering techniques such as bilinear, trilinear, and FSAA, which are very important to cleaning up overall image quality, AF looks more like an expendable luxury. We didn't notice any real image quality differences between varying AF levels in our testing. Our sample screen shots, for instance, looked exactly the same.

Update: Well, actually, they don't. AF can make a discernible difference in certain kinds of games, with first-person shooters being a good example. And if those improvements can be had without killing off too much frame rate, then AF is something you'll want to use. But, if you want FSAA and AF at 1600x1200, Radeon 9700 will pretty much take you there, whereas GeForce 4 Ti 4600 really starts to wheeze when these kinds of demands are put on it.

This article is a joke, the updated screen shot for SS SE shows a huge improvement by my standards using AF. Using AF in flight sims is a god send for ground textures from my experience. Really seems like the author messed up the initial testing of AF and was totally inexperienced with using AF from the get go. I also believe the Radeon9700 has no problem doing Tri-linear samples and AF in openGL. The program just has to be set to tri-linear with the drivers in quality mode (I could be wrong but saw it here on the board).

 

[Ailuros]

Quincunx AA is a BLUR filter..nothing more and IMO should not be classed as FSAA, you can get the same effect by buying a pair of glasses and rub baby oil on the lenses when playing or a cheap Komodo monitor...and screenshots are not capturing that effect as proven by Sharkfood, so don't go there.

Actually it´s 2xRGMS with a 5-tap gaussian blur filter on top. In cases where a game has massive aliasing (SS:SE is a good example for that) and you can´t really use an alternative sampling method due to performance restrictions, with LOD offset to -1.0 the result is slightly more blurry than plain 2xRGMS, while reducing by a fair amount texture aliasing.

As far as screenshots concerns the inaccuracy is more related to text (and that not as a general rule, since at times you can get some type of text blured in a screenshot with some other appearing crystal clear) from what I´ve seen than the actual scenery. If I´d leave everything on default and capture a 2xRGMS and a Quincunx screenshot you CAN tell the difference between the two.

Randell,

I meant 2xSSAA compared to 16-tap anisotropic.

 

[Xmas]

This is incorrect, this was discussed before on this forum and ATI does it per pixel not per polygon. The AF requirements will vary in a polygon. Now Nvidia may do per polygon but ATI doesn't. This could very well be the reason why there is less performance loss with the ATI method over Nvidia's.

Both methods are, without a trace of doubt, calculated per pixel (or maybe for every other pixel), not per polygon.

 

[demalion]

Now that we have Mr. Salvador here, I wonder if he'd be willing to share with us where he got the information on the techniques that he put in the article?

It still reads to me like he was spoon fed "technical info" from one company , and if this is not the case this would be a good opportunity to refute that directly and in detail.

 

[Hellbinder]

You'd be hard pressed to notice any difference while playing the game...

Alexsox, that and several more more statements you have made in this thread are completely and utterly of questionable integrity. I don't know if I am more irritated with your obvious shaded commentary or the regulars here who are saying its *just an opinion*. There comes a point where IHV worship needs to take a back seat to what everyone can see. And I am not just talking ATI stuff here. I also own a GF4, Aniso on both makes a HUGE difference in games, and yes in motion to (i cant believe that I actually have to point that out).

Just becuase (currently) Ati has a better Aniso solution.... i mean really... What will your opinion be when the Nv30 debutes? As I am sure its Aniso speed will be much improved.

 

[Hellbinder]

I'm still of the opinion that in most cases, if you have to choose between AF or FSAA, I'll take the latter, though as someone pointed out on this thread (and I said as much in my piece), Radeon 9700 often doesn't make you decide which you want, since you can often get both and playable frame rates.

FSAA does indeed make a noticable difference, especially at 4x mode. However SO DOES AF at high levels like 8x and 16x. I simply can not see, based on ONE FLIPPING GAME that you can come to a conclusions like you do. I play a multitude of games and AF makes a huge difference in the overall Quality. Flight sims, RPG's and Sports titles see a marked difference in quality. There simply is no comparrison. It just kills me that your entire article seems geared toward selling everyone the idea that *its ok that Nvidia cant do AF*.. because *its pretty useless anyway*..

There is no way you could ever convince me that you did not have an *agenda* before you started.

And BTW the 9700 lets you use FSAA with high levels of aniso in EVERY game currently available. With the exception of no FSAA in pure 16bit titles.

 

[Geo]

Anybody notice this?:

"nVidia contacted us a few weeks ago, and accused ATI of possible AF corner-cutting. At the same time, nVidia product marketing management told us that the GeForce Ti actually applied AF to every pixel on the screen. Subsequently, nVidia engineers contradicted that explanation.
After a lot of back and forth, we finally got to the bottom of the mess. It turns out that nVidia also uses an adaptive implementation, although nVidia's AF uses trilinear filtering plus AF (equivalent to ATI's "Quality" setting for AF). The differences between ATI's and nVidia's implementations turn out to be more subtle than we first suspected, or were first told by nVidia."

 

[alexsok]

Alexsok, that and several more more statements you have made in this thread are completely and utterly of questionable integrity. I don't know if I am more irritated with your obvious shaded commentary or the regulars here who are saying its *just an opinion*. There comes a point where IHV worship needs to take a back seat to what everyone can see. And I am not just talking ATI stuff here. I also own a GF4, Aniso on both makes a HUGE difference in games, and yes in motion to (i cant believe that I actually have to point that out).

Just becuase (currently) Ati has a better Aniso solution.... i mean really... What will your opinion be when the Nv30 debutes? As I am sure its Aniso speed will be much improved.

Would you mind STOP taking my words and twisting them for your own needs?!

As I said many times, that's my own opinion.

I praised ATI already that they offer the BEST AA quality and the BEST aniso quality, all with a very small perfomance hit, which was the reason I noted that the ONLY card that can provide you with a fluid experience with both of these features enabled and set to the max is the Radeon9700pro.

If NV30 comes out and it'll be on par or better - good for nVidia, why should I care?! What's IHV favoring has anything to do with that? Christ Hellbinder, STOP twisting my words!

As far as I'm concerned, I prefer to have both enabled, as one accounts for the faults of the other, but if I had to choose between them, I'd choose AA. Just my 2 cents.

 

[mboeller]

to stir up the waters a little bit more; I post here a few pics made with XMas "polygon tunnel" using the NV25 and the R300 to illustrate the differences between the different AF implementations.


The png-impages were made and posted by Quasar at the technology-forum @ www.3dcenter.de and show the mipmap-levels of different AF settings and implementations :


upper half NV25; lower half R300 :

Links :
http://www.forum-3dcenter.net/vbull...6b711190aecbe6fe33c5809fe504b2&threadid=34199

http://www.forum-3dcenter.net/vbull...32136&perpage=20&highlight=R9700&pagenumber=1

 

[Thowllly]

That confirms the conclusion that was reached in another thread, that the R300 is better than GF4 at 90 and 45 degree angles, but worse at 22 degree angles...

Where can I download this test? I want to test my GF2s 2x ansi

 

[Thowllly]

Hey! What this? I've waited several minutes, and still nobody has posted a link to that tunnel test!

In the mean time, I'll include some obvious observations I've made:

GF4:
1. AF level is selected on a 2x2 pixel level. Not surprising, since it has 4 pipelines. My GF2 does the same.

2. Polygon edges (I'm assuming it's polygon edges that makes those spokes appear) is handled pretty well.

3. Around 45 degrees, most of the time a higher level of AF is used, except between mip level 0 and 1, where a lower level of AF is used. This is interesting, because my GF2 also has some weirdness between MIP level 0 and 1 when using AF (It's only visible at low resolutions, though) Edit: I'm talking about the 8x shot here. At 2x, the AF level seems to be higher for all mipmap levels at 45 degrees.


R300:
1. AF level is selected on a 2x2 pixel level. That is a bit surprising, considering the R300s 8 pipelines. Are they arranged as 2 independent groups of 2x2 pipes?

2. Polygon edges are not handled very well at all. Would be interesting to see the same image but with fever polygons to get fewer distracting lines in the picture.

3. At 45 degrees the level of AF is almost as high as at 90 degrees, but not quite. At 22 degrees it AF level seems to drop to 2x.


That's it. I've draw no conclusions from my extensive research

 

[noko]

I don't understand what the test is showing, one looks AA, the GF4 and the R300 doesn't. Is the Radeon9700 doing Tri as well or is it Bi-linear in the test? Anyways I too would like to run this test on my GF3 to compare.

 

[Thowllly]

The test shows what mipmap level is used at different angles (or maybe I should say, what lod bias used at different angles). The closer the colors are to the centre, the better. The black and white rings are there to make it easier to see how far the away the mipmaps have been pushed.

In my previous post, I should have said 'mipmap level' instead of 'AF level', since the test only shows what mipmap level is used, it doesn't really show what AF level is used. If the AF level isn't high enough compared to how far away the mipmaps are pushed, you will get aliasing.

I don't think any of the shots use AA, but one uses lossless compression (.png), and the other does not. Both cards have trilinear filtering enabled.

 

[Xmas]

Where can I download this test? I want to test my GF2s 2x ansi

Currently you can't, for two reasons. First, I don't have access to webspace right now (which is not really a problem ), and second, I don't want to make it available in its current state. It's neither complete, documented nor very intuitive, so I'll rework it when I find the time. It just leaves too much space for misinterpretation if you don't know what the settings mean.

 

[Xmas]

btw, it's a tunnel with 150 faces, so these lines are really polygon edges.

 

[Thowllly]

Currently you can't, for two reasons. First, I don't have access to webspace right now (which is not really a problem ), and second, I don't want to make it available in its current state. It's neither complete, documented nor very intuitive, so I'll rework it when I find the time. It just leaves too much space for misinterpretation if you don't know what the settings mean.

Oh no! I was so looking forward to playing with this thing! On the other hand, if that test got spread on the internet, we would be flooded with articles claiming that the R300s AF is more buggy than ever, so it's probably for the best

 

[Xmas]

Oh no! I was so looking forward to playing with this thing! On the other hand, if that test got spread on the internet, we would be flooded with articles claiming that the R300s AF is more buggy than ever, so it's probably for the best

You haven't seen the R200 shots yet

 

[Doomtrooper]

You haven't seen the R200 shots yet

Make the program available to the public, I'd like ATI to look at it.

 

[dave_salvator]

Now that we have Mr. Salvador here, I wonder if he'd be willing to share with us where he got the information on the techniques that he put in the article?

It still reads to me like he was spoon fed "technical info" from one company , and if this is not the case this would be a good opportunity to refute that directly and in detail.

The technical info in the article comes from both companies. The long quote from Kirk was from an email I got from him, and I put it into the article because it was a good succint explanation of how nVidia is doing AF. I spent a lot of time talking with both companies about their respective implementations.

As for bringing an agenda to the article, there was no such motivation. I'm not excusing nVidia for taking a larger performance hit than ATI when doing AF, and I do say in the article that the R9700 will almost never force you to choose between FSAA or AF, even at high resolutions like 16x12. It is clearly the superior performer versus anything nVidia currently has shipping.

Our review of the R9700 bears out my opinion of that GPU (http://www.extremetech.com/article2/0,3973,475978,00.asp). To quote myself:

This time around, the brass ring clearly belongs to ATI. And while nVidia is never one to be counted out, the high-profile GPU maker finds itself in the uncomfortable position of looking squarely at its primary competitor's tail lights. And its new can of nitrous oxide that will be NV30 is still some months away. Even if NV30 ships before year's end, nVidia will have missed the all-important Q4 holiday buying season with its next flagship GPU. But you can't summarily dismiss GeForce 4 Ti 4600 now that Radeon 9700 has arrived. While the Ti 4600 can't go the full fifteen rounds, our data shows that it keeps frame rates playable in the majority of test cases, and only buckles under the most severe test conditions. Respectable though it may be, there's only one winner in a flat-out performance contest, and Radeon 9700 scores well ahead of GeForce 4 Ti 4600, at times with more than a 2X performance lead.

Many have opined that what we're seeing is essentially the same old song and dance, and that NV30 will "blow Radeon 9700 away" when it gets here. But since no one has any hard benchmark data on NV30, these conjectures may be premature. nVidia has stated publicly that NV30 will be roughly 2-2.5X the performance of GeForce 4 Ti. But as our data have shown, Radeon 9700 is often ahead of the GeForce 4 Ti by that margin, which leads us to consider that Radeon 9700 and NV30 may be more evenly matched than previous top-end offerings from both chipmakers. Add to the mix that ATI has several "kicker" products based on the Radeon 9700 core brewing, and that R9700 has support for DDR-II memory and is still running on 0.15-micron process technology, and you can see that NV30 will have its work cut out for it when it arrives. It begs the question: if NV30 is only 10-15% faster than Radeon 9700, is it worth waiting for? This is admittedly pure speculation on our part, and NV30 might well mop up the floor with Radeon 9700, though given the performance we've seen from ATI this round, that will be a tall order indeed.

I did make a mistake in getting my Serious Sam screenshots for the article, and I've admitted as much. I also updated the article to say that I found another instance where AF was making a difference, albeit a more subtle one. But, if both AF and FSAA are going to drag my frame rate down to an unacceptable level, then I'll jettison AF first, since FSAA helps clean up texture crawling and line jaggies, whereas AF only addresses texture quality. It's not a decree, it's an opinion, so please take it as such.

Cheers,

Dave
ExtremeTech