Radeon 8500 Aniso vs Geforce 4 Aniso

[Dave Baumann]

The I found the following quote by Tony Tamasi from Bens nVIDIA E3 report a little interesting:

ATI only anisos about 10% of the given screen at a time, while Nvidia does the whole scene

I've done a very quick analysis of IQ and performance on Radeon 8500 and GeForce 4 and I'd like to hear people thoughts in relation to the above quote.

The analysis can be found here.

 

[Tahir2]

I dont know what to say but the Radeon 8500 image with Aniso on looks sharper and better than either of the two Geforce 4 Aniso images... are you sure you haven't muddled them up?

Honestly I am not trying to big up the Radeon 8500 because I own one.

Would love to hear what other people think!

 

[muted]

Switching back and forth from the R8500 and Gf4ti Tri aniso

the R8500 quality is by far superior , and the perfomances are superior when Aniso is enabled ....

I don't see where this 10% comes to hurt ATI .. if it's faster and looks better than the gf4ti .. and it is


I'm curious to see if anyone can make any sense of the 10% comment

 

[Ichneumon]

Misae, I was just about to post the same thing.

Judging only from those 3 shots, IMO the 8500 shot looks better than both the GF4 shots as well.

The walkway is sharper, the tops of the building on the right and left of the screen are sharper, and there is blur on the textures at the end of the walkway in the GF pics i don't see in the 8500 shot. The trees also look a lot better in the 8500 shot... More defined.

 

[Rambler]

Oh well, maybe my eyes are not in as good condition as few years ago, but i've looked at these pics (only radeon and gf4 tri aniso to be exact) and i had to look very closely to barely see any difference - imho radeon looks little sharper on the building on the right, but gf4 ti looks little smoother on the right arc where there's the blue strip. and by little i mean really tiny bit, because i had to switch between the images few times to notice any difference at all.

anyway if you call radeon much sharper on these pics that gf4 then you must have much sharper sight then me i guess.
to be honest i think that in any moving pictures (in-game/video/whatever) you just can't say this image is sharper than the other one

 

[tobbe]

The I found the following quote by Tony Tamasi from Bens nVIDIA E3 report a little interesting:

ATI only anisos about 10% of the given screen at a time, while Nvidia does the whole scene

I've done a very quick analysis of IQ and performance on Radeon 8500 and GeForce 4 and I'd like to hear people thoughts in relation to the above quote.

Theres not much to say afaics, as the quoted guy obviously has no idea what he is talking about. Everyone knows that typically only a small portion of screen area has high degree of anisotropy, so you dont need to "aniso" every pixel.

As for the images in the analysis, the Radeon image looks a bit sharper in places but also more aliased (look at the black arcs in the middle, the Geforce image almost looks antialiased in comparison).
Looks like its just as much a comparison between texture LOD (or LOD selection accuracy) as a comparison between anisotropic filtering.

 

[Reverend]

Maybe a flight sim where you're flying fairly low/near to the ground and your plane tends to roll a bit may give a new perspective.

 

[noko]

Dave,
On your performance graph for the Radeon8500 the graph is labelled 16x while in the data section it is recorded as 8x. Which was it? I know it was the fillrate test in the graph, could you clarify.

Also if you use the ExtremeGFX Add-on did you disable TRUFORM on the Radeon8500? Reason why I mention it is because the GF4 isn't able to do N-Patches which would penalize the Radeon8500 in your benchmark.

The performance hit on the GF4 was monsteous allowing the Radeon8500 to sweep past or catch up to the Ti4600, highly disappointing to say the least.

Also the still image of the Radeon8500 looks better or have more detailed textures from my viewpoint. Not convince it would be better on the Radeon8500 since in motion the Radeon8500 could have gross texture aliasing.

The 10% thinking isn't supported in the images you provided, the walkway itself exceeds 10% of the image, which is anisotropically positioned to our view.

Yes, Reverend does have a good point with a flight simulator, still when I fly I am mostly flying straight or near straight otherwise I wouldn't get to far .

 

[Doomtrooper]

My opinion on this, and my reply to Tamasi... Horse Poop..
Radeon 8500

Geforce 4

As for Rev's comment on Flight Sims..thats a vailid point yet I know it's NOT 10% like Tamasi stated... and that' s what the question is about

 

[noko]

Looking at the buildings in the background the Radeon8500 does the roofs alot better on the sides which you may want to emphasize in your marked up image.

 

[Doomtrooper]

Looking at the buildings in the background the Radeon8500 does the roofs alot better on the sides which you may want to emphasize in your marked up image.

Oops tunnel vision, coming from Pro-Ut player...look straight ahead.. what's in front of you

 

[noko]

What is funny is this is the problem I have, I can only use 4x or 32tap on my GF3 running at greater then Ti500 speeds. So being able to do 8x in the first place on my Nvidia card would be a significant improvement for me. Now looking at the Radeon8500 images, my eyes just water and really makes what I play with seem bad .

 

[DemoCoder]

#1 the 8500 images look more aliased on the far left/right.
#2 Why is the standing grass on the right missing in the 8500 screen?
#3 These tests are invalid until we see them compared in motion
#4 3D graphics is all about the art of cheating. Most people can't tell the subtle difference. While ATI's algorithm may be "less correct", no one is going to tell the difference, I bet, in the vast majority of game situations.

People on this board agonize over blown up still shots of various 3D scenes trying to argue over small pixel areas and translate those into comments like "WAY SUPERIOR" The average person, if seeing these two games running side by side on two monitors, would not experience a "night and day" effect, in fact, they may be hard pressed to tell the difference at all.


Obviously, the rasterization engines should adapt their texture filtering and AA sampling to those areas which need them most, and they should endeavor to be "correct" most of the time and avoid artifacts by making the logic smart.

My opinion is that there isn't a big enough difference to worry about it. Any form of anisotropic is better than plain old bi/trilinear. Score 1 for ATI. On the other hand, ATI doesn't do multisampling and doesn't adjust AA to where it is needed most. Score 1 for NVidia.

Obviously, the ideal next-gen card would do the optimal thing for both texture filtering and FSAA. Let's have 128-tap filtering and 16X AA on those texels and pixels which need them most and in the optimal pattern for that particular triangle, and that particular texture map.

 

[Doomtrooper]

Is there more than 10% of the screen recieving Ansitropic filtering ....that is the question.

 

[noko]

LOL Doom

Democoder,

For some it is more obvious the differences and we do notice the improvement or maybe I should say preference of one image over another. The walkway would be immediately noticeable at least by me while the background would make a more subtle impression but noticeable. Why noticeable? Because there is a mismatch in texture detail between the surfaces facing you and the surfaces which are anisotropically positioned to your view. In other words the front of the buildings have sharp textures and the sides are blurred out (a mismatch) on the Ti4600. The Radeon8500 renders the sides of the builders better more consistent with what the front of the building shows, which makes the buildings more realistic to me.

 

[Jerry Cornelius]

It would be nice of there was a reference scene textured purely using radiosity (right word?) algorythms. Then different cards could be compared bitwise against it to see how close their filtering matches.

It's all just a hack to get around actually blending every element of the artwork represented by whichever pixel.

I think that just two or three bilinear samples, taken from different mip levels depending on the polyygon, would be rationally as accurate any any other scheme. In fact, I'd be curious to see how good trilinear would look with a bit of anisotropy offsetting the two texels.

Obviously 64 samples can more accurately describe the contents of what any given pixel is reproducing at a very high cost. But stuff in the distance isn't likely to look very different and up close textures aren't detailed enough yet to be described with more than a couple of samples. So it may be just the stuff in the middle that is going to benefit.

EDIT
I wouldn't be surprised if NVIDIA shots don't look as sharp because the texturing up close is oversampled. Oddly enough, in those screen shots it's the brick in the bottom of the left circle that stands out to me. I would have thought that NVIDIA shot would be better. They must not be weighting the samples accurately.

 

[Crusher]

Hrmm... I can't see any noticable difference in the areas you've circled. In fact, the only area I can notice a discrepancy between the two screenshots, is with the environment map of the sky being reflected on the black arches on either side. To me, it looks much smoother on the GF4 screenshot than on the Radeon. Apart from that, I wouldn't be able to tell them apart, but maybe that's just me.

 

[Doomtrooper]

Crusher,

Well thats good because Tamasi is saying that ATI is only doing 10% of the screen hence the lower performance hit the 8500 takes doing anistropic...I notice a big difference between the two shots, I use a program called Acdsee and compare them side by side easily..

 

[KimB]

I just have to say that the Radeon shot is full of texture aliasing. This texture aliasing will be far more visible when in motion.

Personally, I can't stand texture aliasing in the least.

The question is: Why is there lots of texture aliasing in the ATI shot?

There are two possibilities that I can think of:
1. ATI chose very aggressive LOD for the textures.
2. ATI's aniso algorithm doesn't use enough texture samples.

Given the very small performance hit of ATI's aniso, I suspect number 2 is the case. My main evidence for this is that even with 16x aniso, the Radeon 8500 has very little performance hit.

Besides, I'm not sure that just having too aggressive LOD would cause the jagged look that the Radeon 8500 shots seem to be showing.

 

[Doomtrooper]

I just have to say that the Radeon shot is full of texture aliasing. This texture aliasing will be far more visible when in motion.

Personally, I can't stand texture aliasing in the least.

Why did I know you would say that