R420 IQ Comparison

[jb]

You know what would have made a far better comparision a long corridor flat corridor such which would be used to examine AF.

Dave has such a UT map... However the guy that made it put the spawn point outside the world, so "bad" things happen when he tried to spawn. What a dumb@ss that guy was...oh wait a tick..I will fix that map today and get it back to Dave as it was my fault in the first place.

 

[WaltC]

You missed a part from that review though:

The interesting thing is that X800 and Refrast look better on the screenshots in a head to head comparison. This is because of the higher level Mipmaps at 45-degree angles. Although this looks better on screenshots, it can cause sparkle in motion.

Here´s what Microsoft has to say about it:

Quote: "The DX9 reference rasterizer does not produce an ideal result for level-of-detail computation for isotropic filtering, the algorithm used in NV40 produces a higher quality result. Remember, our API is constantly evolving as is graphics hardware, we will continue to improve all aspects of our API including the reference rasterizer."

Generally, when quoting someone directly, it is common practice to name the individual quoted--or, if you are referencing a printed quote you found elsewhere, to attribute the quote with a link to the source.

In this case we have no attribution whatever for this quote as it is reprinted here by THG, and in any case a mere assertion of a quote without attribution is not considered reliable journalistically, and should always be viewed with frank skepticism anytime it is encountered.

Additionally, the part of the first sentence, "the algorithm used in NV40 produces a higher quality result" is suspicious for a couple of reasons. First, it is grammatically poor and appears as more of a run-on sentence which has merely been added to first sentence in the quote, and second, the context of "the algorithm used in NV40 produces a higher quality result" is wrong for the quote itself, but right for the context of THG article in which it appears. This may indeed be a genuine M$ quote from someone, on the general topic of the reference rasterizer, but I suspect that the nV40-specific line has been inserted by THG--which neatly explains why the quote is not attributed to anyone in any manner, except "Microsoft," which does not inspire confidence in its authenticity. If we remove the phrase "the algorithm used in NV40 produces a higher quality result," in fact, then the quote becomes something much more likely to have been stated by an individual employed by M$ in the context of the DX reference rasterizer.

Note, specifically, that nothing in the "Microsoft" quote has anything whatever to do with the concept of "causing sparkle in motion," yet this is presumably the "it" which THG represents the M$ statement as pertaining to.

 

[Quitch]

I took it more as a sign of just how desperate people are to portray this in a negative light, and it has taken all this time for all these threads to finally start reaching the conclusion that, in actual fact, this method may well look better than the naive trilinear method.

There's more testing to be done, but I hope we're past the full on knee jerk phase now, that was just irritating.

It's not that difficult to understand that people are upset in the light of Ati's PR documents talking about optimizations and the same workload when doing filtering + that you should use color mipmapping tools to check to check what the card is actually doing.

And claiming that they've passed the WHQL tests as a proof of IQ is hardly making things better.

AFAIK the Quality button is still labelled Quality, and the card still does trilinear when requested to do trilinear... or is this another case of people renaming the Quality button in their minds?

 

[Lezmaka]

You missed a part from that review though:

The interesting thing is that X800 and Refrast look better on the screenshots in a head to head comparison. This is because of the higher level Mipmaps at 45-degree angles. Although this looks better on screenshots, it can cause sparkle in motion.

Here´s what Microsoft has to say about it:

Quote: "The DX9 reference rasterizer does not produce an ideal result for level-of-detail computation for isotropic filtering, the algorithm used in NV40 produces a higher quality result. Remember, our API is constantly evolving as is graphics hardware, we will continue to improve all aspects of our API including the reference rasterizer."

Generally, when quoting someone directly, it is common practice to name the individual quoted--or, if you are referencing a printed quote you found elsewhere, to attribute the quote with a link to the source.

In this case we have no attribution whatever for this quote as it is reprinted here by THG, and in any case a mere assertion of a quote without attribution is not considered reliable journalistically, and should always be viewed with frank skepticism anytime it is encountered.

Additionally, the part of the first sentence, "the algorithm used in NV40 produces a higher quality result" is suspicious for a couple of reasons. First, it is grammatically poor and appears as more of a run-on sentence which has merely been added to first sentence in the quote, and second, the context of "the algorithm used in NV40 produces a higher quality result" is wrong for the quote itself, but right for the context of THG article in which it appears. This may indeed be a genuine M$ quote from someone, on the general topic of the reference rasterizer, but I suspect that the nV40-specific line has been inserted by THG--which neatly explains why the quote is not attributed to anyone in any manner, except "Microsoft," which does not inspire confidence in its authenticity. If we remove the phrase "the algorithm used in NV40 produces a higher quality result," in fact, then the quote becomes something much more likely to have been stated by an individual employed by M$ in the context of the DX reference rasterizer.

Note, specifically, that nothing in the "Microsoft" quote has anything whatever to do with the concept of "causing sparkle in motion," yet this is presumably the "it" which THG represents the M$ statement as pertaining to.

The quote is from this site:

http://www.techreport.com/onearticle.x/6668

 

[WaltC]

The quote is from this site:

http://www.techreport.com/onearticle.x/6668

Well, thanks for illustrating the point--Heh... (TR also quoted "Microsoft" without naming the individual responsible for the quote...this is a very poor journalistic habit--much better to say, "John Doe, an employee [job title] within M$'s DX software division, said, "...".)

But, clearly then the THG context for the remark is wholly inappropriate, as the TR article from which THG clipped the quote has nothing to do with ATi's drivers and hardware at all, but deals only with nVidia's drivers and nVidia's hardware--which is the focus of the TR article--the 60.72 Forcenators. The context of the comments in the quote is one of actually saying the nV40's algorithm is better than the DX9 rasterizer output for nV40.

But THG yanked the unattributed TR quote from a M$ employee about the 60.72 nVidia drivers, and nothing else, and specifically misused that quote in the THG article to make it appear as if M$ was talking about the same IQ comparisons between R4x0 and nV40 THG was talking about, when M$ was talking about no such thing at all. Thanks for clearing this up as it proves THG quoted the unknown M$ employee completely out of context.

In fact, since the M$ employee is saying that the nV40 algorithm is better than the DX9 rasterizer output for nV40, then might not the same be true for R4x0? Certainly seems likely, and if so, then the entire premise THG makes as to "sparkle" and whatnot is undercut by the real context of the unknown M$ employee's remarks as first quoted by TR, and then clipped for use by THG.

My guess is that if the THG article had included a link to the TR article, then it would have been obvious the THG article was using it out of context, and so no attribution as to the source of the quote THG used was provided.

 

[Ostsol]

(TR also quoted "Microsoft" without naming the individual responsible for the quote...this is a very poor journalistic habit--much better to say, "John Doe, an employee [job title] within M$'s DX software division, said, "...".)

Sounds like the Inq. . .

 

[WaltC]

(TR also quoted "Microsoft" without naming the individual responsible for the quote...this is a very poor journalistic habit--much better to say, "John Doe, an employee [job title] within M$'s DX software division, said, "...".)

Sounds like the Inq. . .

Oh, yea!... Heh... I didn't mean to say "John Doe," literally. I meant for them to use the person's real name, and was only using "John Doe" as a placeholder for an example of the format. Heh...But I know what you mean... Too many John Doe's and John Smith's out there already

 

[Pete]

16AF in Halo with normal Ati AF only applying trilinear to the top texture:
http://www.gaeugf.ch/ted/halo2.jpg

16AF in Halo with trilinear forced through all stages with a 3rd party application:
http://www.gaeugf.ch/ted/halo1.jpg

Note the wall to the left, draw your own conclusions.

Is this representing quality degradation of ati's new filtering feature or another "mistakenly" grabbed shot?

I suppose that zig-zagging may be meant to disappear with trilinear filtering, but the issue of tri on all texture stages is separate to "fast trilinear." BTW, if you leave AF set to App Pref in the CP and enable it in-game, you'll get tri on all texture stages. This has been well-documented: ATi deems it nec'y to only apply tri to the first texture layer when you enabled Quality AF via their CP; Perf AF means bi all the way.

 

[jimmyjames123]

16AF in Halo with normal Ati AF only applying trilinear to the top texture:
http://www.gaeugf.ch/ted/halo2.jpg

16AF in Halo with trilinear forced through all stages with a 3rd party application:
http://www.gaeugf.ch/ted/halo1.jpg

Note the wall to the left, draw your own conclusions.

Not only the left wall, but the gun in the halo1 pic seems to look better too.

 

[Pete]

The gun looks exactly the same to me...?

 

[trinibwoy]

The gun looks exactly the same to me...?

Gun looks the same to me too but I'm half blind

Left wall does look different though.

 

[Evildeus]

The gun looks exactly the same to me...?

Gun looks the same to me too but I'm half blind

Left wall does look different though.

The guns are different, now which one is the good one, is another question. The differencies i can spot:
- on the top, 2 differencies (the black grey line and the far end of the black grey area)
- at the end on the left of the gun, the circle.
*edit* but the position is not exactly the same, so i don't know the reason of the differencies.

 

[Blastman]

16AF in Halo with normal Ati AF only applying trilinear to the top texture:
http://www.gaeugf.ch/ted/halo2.jpg

16AF in Halo with trilinear forced through all stages with a 3rd party application:
http://www.gaeugf.ch/ted/halo1.jpg

Note the wall to the left, draw your own conclusions.

Not only the left wall, but the gun in the halo1 pic seems to look better too.

That doesn’t look like a mipmap transition. It looks more like a graphical glitch. The 2 pictures are not taken from exactly the same spot. I’ve seen situations in games where in certain spots where if you just turn slightly one way you can get a “graphical glitch” … ie. … a section of graphics that turns colors, turn slightly back and it goes away. Those shots need to be from exactly the same spot.

 

[jimmyjames123]

The gun looks exactly the same to me...?

Are you guys looking at the same pics that I am?

Open them up full size window and flip back and forth. The light gray triangular portion at the top that looks different; the black lining at the top of the gun around the dark gray area looks different; the knob at the front of the gun, left side, looks different; even those circular yellow objects look different on the back of the gun. The halo1 pic looks better to me.

EDIT: true, it would be nice to see them in the exact same spot.

 

[muzz]

Explaining trilinear filtering to you would be a waste of time. Assuming you think the NV17 image is correct and the X800 is wrong, I want to know how you came to that determination. It should be simple for someone that already has all the answers.

Don't assume anything with me, I merely pointed out what I recognize as a noticible difference. If you can't see the difference in the screen shots, you're either blind, tired or lying to yourself.

Are you referring to the red circled screenies?
If so I can see a difference, it is slight and the ATi looks a bit sharper.
On the entrance wall as well.
As stated these shots are not the best, as the depth is tough to judge IMO.
Those red circles are pretty darn close, Edit: I am wondering if it is the lighting that is causing it ( notice the sky).

 

[muzz]

I also like the pictures that noko posted better, it is longer, lighter and much easier to see things.
If someone used that shot I believe it would be easier to see.

Great shot noko.

 

[Xmas]

But, clearly then the THG context for the remark is wholly inappropriate, as the TR article from which THG clipped the quote has nothing to do with ATi's drivers and hardware at all, but deals only with nVidia's drivers and nVidia's hardware--which is the focus of the TR article--the 60.72 Forcenators. The context of the comments in the quote is one of actually saying the nV40's algorithm is better than the DX9 rasterizer output for nV40.

But THG yanked the unattributed TR quote from a M$ employee about the 60.72 nVidia drivers, and nothing else, and specifically misused that quote in the THG article to make it appear as if M$ was talking about the same IQ comparisons between R4x0 and nV40 THG was talking about, when M$ was talking about no such thing at all. Thanks for clearing this up as it proves THG quoted the unknown M$ employee completely out of context.

In fact, since the M$ employee is saying that the nV40 algorithm is better than the DX9 rasterizer output for nV40, then might not the same be true for R4x0? Certainly seems likely, and if so, then the entire premise THG makes as to "sparkle" and whatnot is undercut by the real context of the unknown M$ employee's remarks as first quoted by TR, and then clipped for use by THG.

I don't quite understand what you mean. Microsoft says that the LOD selection algorithm used in NV40 is better than the one used in the reference rasterizer. NV40 uses a pretty accurate representation of a formula that is widely accepted as the "correct" one, as can be seen in the OpenGL spec. From the images provided we can also see that R420 (actually most other GPUs) uses the same algorithm as the reference rasterizer.
If A > B and B = C, then A > C.

 

[noko]

Thanks muzz,

Now if someone with a X800 or even a RV360 would expound on this using Mafia I think would be most helpful. Interesting is that you can tell the mipmaps and the blending being done with Trilinear right in front of your eyes with this game. I don't think there will be any hiding or magnifying glass requirements to see if it works or not. Any takers?

 

[Maintank]

That's my point. Just what do you use to determine which image is correct? The only thing we have is the reference raster which it seems ATI is very close to. How can it be said that ATI is cheating when the image is so close to the reference?

And reading through this my opinion is sumed up pretty good right here.

It appears there is no standard for filtering methods that anybody can point and say "Hey this is how it is technically supposed to look like".

here we have people comparing screenies to microsofts rasterized version and then Microsoft says their version isnt even as nice as an IHVs version.

I would have guessed the rasterized version by Microsoft is as good as it gets.

Apparently I guessed wrong. So, do you think in the future somebody can lay down some standards that IHVs need to adhere to for filtering methods?

 

[Stryyder]

Good point..