ATi in for rough times from some reviewers? Aniso on R300.

[Hellbinder]

There is a nice educational presentation up at the ATi site. IT has a lot of nice information covering the main areas of the R300's tech. In the section covering Smoothvision 2.0 they talk about Anisotropic filtering. I found this....

All modes use an adaptive algorithm that only takes multiple texture samples from the parts of an image that would actually benefit.

While with trilinear filtering will correct much of the complaints about the 8500's Aniso method. I can see Sites like Toms etc refusing to bench the 9700 agaist any of the Nvidia offerings with aniso. As in their mind ATi is cheating. Hmmm.. wonder how this will pan out.

There is a lot of other cool info. And I learned some new things about the 9700. check it out.

http://www.ati.com/vortal/r300/educational/main.html

 

[KimB]

Any true anisotropic implementation is fully adaptive. That is the nature of the method. Isotropic images (viewed face-on) should always use plain bilinear/trilinear filtering, whether or not anisotropic is enabled.

There will be no complaints about the 9700 "cheating" if it produces the same image quality as seen in the GeForce3/4 and Parhelia implementations.

 

[Hellbinder]

i promise not to turn this thread into a flame war to the best of my ability.. However.. I just cant let this go..

There will be no complaints about the 9700 "cheating" if it produces the same image quality as seen in the GeForce3/4 and Parhelia implementations.

Iq has been the same on most surfaces and most game exmples. As all of the first 50 or so reviews including toms origional point out. There was no noticable difference. Until They actually started looking for a problem.
At any rate, At most a reviewer should simply have pointed out the differences and still shown the numbers. PArhelia Aniso has been shown to have very *shoddy* looking aniso. So I dont see where the IQ argument has any merrit at all.

Of course The entire reason you are saying that *adaptive* is ok now is due to Nvidia's latest *driver* changes. I could easily find multiple quotes from you stating that Full screen in the *right way*. You have only changed you mind in the last week. i find it interesting that Tom and others claimed that Ati was doing a *driver hack* and it is *unclear what they are doing with their driver*.. When in fact is has been a hardware algorithm the whole time. Yet Nvidia just releases a *driver hack* to improve performance. That can only be taken advatage of with a tweak program.. yet no mention of *cheating* at all from anyone...

Just a sign of the times i guess..

 

[OpenGL guy]

There will be no complaints about the 9700 "cheating" if it produces the same image quality as seen in the GeForce3/4 and Parhelia implementations.

Why should the image quality of GeForce or Parhelia be assumed to be "perfect"? Different doesn't imply incorrect.

Unless you think that the "better" (very subjective, BTW) result is the one that should be used as a reference. And I don't think that'll be very fair to the GeForce when comparing AA quality

 

[Bigus Dickus]

Initial indication are that ATi has corrected the adaptive filtering algorithm to correctly filter surfaces at rotated angles. I expect the 9700 to maintain the speed of the 8500's previous method, while "matching" that of Matrox and nVIDIA.

 

[KimB]

Why should the image quality of GeForce or Parhelia be assumed to be "perfect"? Different doesn't imply incorrect.

It's not necessarily perfect, but from everything I've seen, neither card has a preference for any angle of surface, and I doubt you could get better image quality without either using brute force (Always using more texture samples/pixel) or enabling a higher degree of anisotropic.

Personally, I never claimed that the Radeon 8500's anisotropic was cheating. I just think that it's pointless to look at the high performance without also considering the poor image quality in certain scenarios. And, as I've repeatedly stated, just a few scenarios is good enough. If it happens 1% of the time, but is noticeable when it does happen, than that's certainly good enough to turn me off to the card.

As for the many reviewers that saw no problems early-on, reviewers often don't actually play with the cards enough, and the problems may not be obvious right away.

And if you still want to consider MSAA as looking poor in a few scenarios, I've already stated that alpha blends are being used. I enabled them in UT (the one game that I play often that used alpha tests), and I'd be highly surprised if any new engine coming out in the near future failed to use alpha blends.

 

[KimB]

Oh, and btw, since each pixel pipeline can apparently filter four bilinear-filtered samples, the performance hit from anisotropic should not be severe on the 9700 (and it doesn't appear to be on the previews).

 

[Bigus Dickus]

Personally, I never claimed that the Radeon 8500's anisotropic was cheating. I just think that it's pointless to look at the high performance without also considering the poor image quality in certain scenarios. And, as I've repeatedly stated, just a few scenarios is good enough. If it happens 1% of the time, but is noticeable when it does happen, than that's certainly good enough to turn me off to the card.

This appears to be corrected with the 9700.

 

[Hellbinder]

Oh, and btw, since each pixel pipeline can apparently filter four bilinear-filtered samples, the performance hit from anisotropic should not be severe on the 9700 (and it doesn't appear to be on the previews).

Except that the previews are using Trilinear filtered samples. Not bilinear. Meaning that Bilinear filtered Ansio will be even faster than what the previews have shown.

 

[Dolemite]

...from everything I've seen, neither card has a preference for any angle of surface, and I doubt you could get better image quality without either using brute force (Always using more texture samples/pixel) or enabling a higher degree of anisotropic.

IMO, the higher degree of anisotropic on the Radeon offers unquestionably better image quality than seen on any Geforce thus far. And yes, I've run them both personally so this is not some f@nboy BS. What I did appreciate about the nVidia implementation is the presence of trilinear in conjunction with aniso, and thus the lack of mipmap boundaries which are still faintly present with ATI aniso. I cannot recall situations where the adaptive nature of the ATI method was noticable or distracting.

Now, as I understand it, the 9700 addresses the trilinear complaint, and IMO, in doing so will virtually eliminate what little distraction the adaptive aniso causes, because any surface that is not anistropically filtered will at least be trilinearily filtered. On those odd-angled surfaces where no aniso is applied on the 8500, the lack of aniso very much increases the visibility of mipmap boundaries. On the 9700, some detail may be lost on these surfaces but no glaring mipmap boundaries will be visible, thanks to trilinear.

So, I'm not sure if the adaptive method and its minor flaws will still be present on the 9700, but it seems that trilinear would effectively "plug the leak" in this method.

 

[KimB]

Which obviously means two trilinear-filtered samples per pixel pipeline.

 

[KimB]

I still hate how people call the 8500's method "the adaptive method."

ALL anisotropic implementations are adaptive. It's the very definition of anisotropic filtering.

 

[demalion]

I still hate how people call the 8500's method "the adaptive method."

ALL anisotropic implementations are adaptive. It's the very definition of anisotropic filtering.

I thought that recent nVidia drivers were achieving performance gains by being more adaptive than in the past? So even though all anisotropic filtering might be adaptive to a degree, until currently (or perhaps still currently) the 8500/7x00 was MORE adaptive?

Or do I misunderstand recent developments?

 

[KimB]

Actually, it appears that the newer drivers use a different MIP map selection algorithm when anisotropic is enabled. It doesn't appear to affect image quality on my machine, but looks kinda funny if you turn r_colorMipLevels on in Q3.

The drivers probably also use a different anisotropic level selection algorithm.

The root of the problem with the GF4's aniso, when enabled, disables the second texturing unit. This reduces performance for most isotropic surfaces.

 

[John Reynolds]

Actually, it appears that the newer drivers use a different MIP map selection algorithm when anisotropic is enabled. It doesn't appear to affect image quality on my machine, but looks kinda funny if you turn r_colorMipLevels on in Q3.

The drivers probably also use a different anisotropic level selection algorithm.

The root of the problem with the GF4's aniso, when enabled, disables the second texturing unit. This reduces performance for most isotropic surfaces.

I think we should wait to see what the major reviewers actually do in their previews/reviews before crying wolf. Yes, there will be, I expect, an unfortunate trend in most stating how NV30 is around the corner and might be the better DX9 part (unfortunate in that I don't think a competitor's product should be mentioned in a review), especially due to Nvidia's recent PR activity, but that's the reviewers' fault.

 

[Althornin]

I still hate how people call the 8500's method "the adaptive method."

ALL anisotropic implementations are adaptive. It's the very definition of anisotropic filtering.

I remember telling you that about 3 months ago on nVnews and getting laughed at by you and your cronies.
Hmm.

 

[KimB]

If you don't believe me, then go look it up.

 

[Althornin]

If you don't believe me, then go look it up.

about what?

 

[KimB]

That all anisotropic filtering is adaptive.

 

[Althornin]

I still hate how people call the 8500's method "the adaptive method."

ALL anisotropic implementations are adaptive. It's the very definition of anisotropic filtering.

I remember telling you that about 3 months ago on nVnews and getting laughed at by you and your cronies.
Hmm.

That all anisotropic filtering is adaptive.

What part of "I told you that three months ago and you laughed at me" dont you get?
I KNOW all aniso is adaptive, i told you months ago, and you laughed at me for it.
There is no need for me to "go look it up".
I KNOW, have known, and knew - when you were still denying it, saying that the GF3/4 didnt do adaptive aniso, etc.
Thats my point. Whats yours?