NV: Everything under control. 512-Fermi may appear someday. Yields aren't under 20%

[MrK6]

Explain the differences between SSAA and MSAA+AF and how that impacts current titles(that is an extremely trivial question).

Isn't that what I asked you to do? In my rudimentary understanding, SSAA simply renders the image at a higher resolution and then downsamples it to whatever the screen size is to get the correct mix of colors. MSAA is an "optimized" form of SSAA that only uses certain values of each pixel, which in turn has it run faster but lack the same quality. Again though, why am I the only one giving useful information if you seem to know so much more?

Personally- I don't like SSAA- and I have *never* made a secret of that. Be that as it may, I do understand why BFG likes it and there really is nothing that can emulate it. The undersampled AF is a much bigger deal to me, and if you think this is me saying this just because of the current matchup- I have an 11 year post history about how obsessive I am on that particular feature(BFG is the same with AA).

OK, why and how so? Summarize it.

If you want to start a thread on signal theory have at it, I'd certainly participate. If you have to look signal theory up, perhaps you should start off reading Foley & Van Damm and get the basics down. The overwhelming majority of people aren't going to care about the differences, there are those of us that do. For those that don't care, enjoying your own ignorance is a much cheaper way to continue with your hobby then getting borderline OCD about IQ.

I think it's funny that you have this elitist attitude about rendering methods, but still haven't said a single, useful phrase here to bring us "into the light." So you either A) don't have the knowledge you proclaim you do, and only walk circles around it by talking ambiguously or B) lack the communication skills to break down your knowledge and present it in precise form that the average user can digest.

Now I'm not saying this to be a dick to you. Look at it this way - I'm a medical doctor by trade (currently a student, but I still see patients). I know a ton about biochemistry, pathology, physiology, and the like regarding all kinds of diseases and ailments that I encounter everyday, 99% of which is over my patients' heads. If I can't break down my knowledge into a form that is useful to them and that they can understand, then I have failed. What I'm pointing out here is that if you say "oh, you ignorant fools, go enjoy your 'video games' while I savor the finer qualities of rendering methods," it's not only obnoxious, but a complete waste of forum space. No one cares how much you think you know, either help others and post information to help the community learn, or don't post at all.

 

[MrK6]

No, you can see the difference by simply testing and playing games. Once you get used to a certain level of image quality, it’s very hard to go back to anything worse because you spot the regressions immediately. Learning the theory just helps explain what you’re seeing.

I’m not allowed to link to my own articles, but Google “ATi 5000 Series Image Quality Analysis” (it’s the first hit). Between that article - and the links to other articles on its first page - you should get a good comparison of the GT200 vs the 4000 vs the 5000 and see what I’m talking about. The differences are quite observable and are easily documented objectively and subjectively. Of course it’s up to each person to decide what they prefer.

It’s also silly to see people accusing Ben of being a focus group member or an nVidia cheerleader, because it’s utter nonsense. He has absolutely no affiliation to nVidia other than being regular consumer of their products. Look up his posts from 1999 and you’ll still see him promoting AF and disliking SSAA.

Thanks for the reading suggestion. I took a quick read through, and I see where the GTX 285's rendering gave a sharper texture. I would ask then, is this only in OpenGL-based games or is it universal? Also, has the 5xxx series AF been tweaked in drivers or is it a limitation of the architecture? I remember AMD touting the 5xxx series angle-independent AF as revolutionary, and while it does seem to work, you do get the excess blurriness (ironic, considering that's why you're running AF in the first place).

My other inquiry would be to see if you think your GTX 470 has an even better rendering method than the GTX285. You said you prefer the GTX 285's method of AF (minimal aliasing with sharper textures) as opposed to AMD's method of AF (no aliasing with more blurry textures), but what has Fermi brought to the table in this regard?

 

[RussianSensation]

It is interesting that AnandTech shows completely different findings from what is reported by BFG and Ben.

http://www.anandtech.com/show/2977/...tx-470-6-months-late-was-it-worth-the-wait-/7#

NV vs. ATI Anti-Aliasing:

"With the exception of NVIDIA’s new TrSS mode, very little has changed. Under DX10 all of the cards produce a very similar image. Furthermore once you reach 4x MSAA, each card producing a near-perfect image. NVIDIA’s new TrSS mode is the only standout for DX10."

Let's look closely at the screenshots of Crysis: Warhead

ATI's AA quality is actually slightly better than NV's. Moreso, to get similar image quality, NV took a bigger hit than ATI with their older generation GTX285 (8x MSAA of 25% vs. 13% for 5870 and 12% for GTX480). Even 4890 was more efficient for 8AA over GTX285.

Image Quality
1. Look at the vegetation at the very top right of the screen.
4890 DX9: 4AA + AAA (adaptive AA) >>> GTX480 32x or even GTX285 DX9: 4AA + TrSS

2. Look at the vegetation right above the gun and the palm tree leaves, as well as the longer end for the leaves being shaded in the top right hand corner.
5870's DX9: 4AA + SSAA >>> GTX480 DX9: 4AA + TrSS.

Problem is, AAA doesn't work in DX10.

Anisotropic Filtering:

"With the Radeon 5870 we saw AMD implement true angle-independent AF and we’ve been wondering whether we would see this from NVIDIA. The answer is no: NVIDIA’s AF quality remains unchanged from the GTX200 series." Stating that GTX285 does AF better than GTX470 is misleading.

"More to the point, we have yet to find a game where the difference between AMD and NVIDIA’s AF modes have been noticeable; so technically AMD’s AF modes are better [With the Radeon 5870 we saw AMD implement true angle-independent AF]"

Clearly AMD's AF is superior (completely rounded circle), but hardly matters in the real world, while AF quality for GTX285 and GTX470 is identical in DX9/10.

 

[BenSkywalker]

In my rudimentary understanding, SSAA simply renders the image at a higher resolution and then downsamples it to whatever the screen size is to get the correct mix of colors.

In the abstract you are correct, though it is a bit more complicated then that. SSAA gives a larger data sample for the coverage area of the pixel to give a more evenly weighted color value- saying it is more accurate is a point of debate for some.

MSAA is an "optimized" form of SSAA

MSAA takes geometric data and creates a higher fidelity "image" of Z values and then compares them locating pixel areas that have varriations in Z value and then samples multiple color samples only from those pixels(in a nutshell, a bit more complicated then that but that gets the general idea across). This is why MSAA falls down with a lot of different effects(alpha textures being the most common and obvious but also shader routines can introduce aliasing as is becoming increasingly obvious with modern games).

OK, why and how so? Summarize it.

In the simplest terms- SSAA gives you a cleaner image; MSAA gives you a sharper image. While it isn't on a technical basis, the essence of SSAA is a blur filter. SSAA can help eliminate noticeable aliasing on anything, at the expense of high contrast levels. I fully understand why BFG likes SSAA, nothing is really comparable to how clean the image is using that method of AA. I personally am a big fan of razor sharp images which is why I am so obsessive about AF quality and why I like MSAA+AF better.

What I'm pointing out here is that if you say "oh, you ignorant fools, go enjoy your 'video games' while I savor the finer qualities of rendering methods," it's not only obnoxious, but a complete waste of forum space.

Oddly enough, those of us that do care already know what the differences are(thanks BFG ). This is why I stated those that don't know what the differences are probably aren't going to care- and for that I was attacked. How many times on this forum do people post things that help people learn anything? The reality is a bunch of flaming fanboys get all pissy if you don't champion their parts here, no matter if you are trying to teach them or not. This thread is yet another example of that. Yes, I do get an elitist attitude about it eventually as their are a small handful of people here that want to learn and a whole bunch that want to cheerlead their company of choice. You are currently a medical student- if we were in a thread talking about taking a certain prescription for a particular ailment I sure as hell wouldn't be bashing you for your take on why one was better then the other, I'd be asking questions to find out why(I understand that is what you are doing now, but look at the other posts in this thread).

My apologies BFG, should have done this the second IQ came up-

http://alienbabeltech.com/main/?p=12648

It is interesting that AnandTech shows completely different findings from what is reported by BFG and Ben.

The point I like to use about AT's IQ analysis is one I have used for many years-

It's important to note that in most cases (such as the one above), you won't be able to tell any difference between ATI's performance and quality anisotropic filtering settings.

http://www.anandtech.com/show/970/14

That is bilinear versus trilinear- tire iron to the forehead obvious and Anand didn't notice it. AT is probably the poorest site I've seen for IQ analysis(I'm assuming it is vision related as missing bi vs tri is so shockingly obvious it is hard to comprehend how anyone could miss it).

 

[RussianSensation]

AT is probably the poorest site I've seen for IQ analysis(I'm assuming it is vision related as missing bi vs tri is so shockingly obvious it is hard to comprehend how anyone could miss it).

I am not sure what the relevance is for linking 9700 Pro image quality article. I am not disputing what the best website is for IQ comparisons.

Just look at the screenshots linked. You can clearly see that ATI anti-aliasing modes produce more detail in the palm tree leaves, as well as the vegetation above the gun, than NV's TrSS modes.

 

[BenSkywalker]

Just look at the screenshots linked. You can clearly see that ATI anti-aliasing modes produce more detail in the palm tree leaves, as well as the vegetation above the gun, than NV's TrSS modes.

Got a link to png? One of the many reasons AT isn't a good choice for IQ comparisons(jpegs for IQ comparisons, heh).

 

[MrK6]

In the abstract you are correct, though it is a bit more complicated then that. SSAA gives a larger data sample for the coverage area of the pixel to give a more evenly weighted color value- saying it is more accurate is a point of debate for some.

MSAA takes geometric data and creates a higher fidelity "image" of Z values and then compares them locating pixel areas that have varriations in Z value and then samples multiple color samples only from those pixels(in a nutshell, a bit more complicated then that but that gets the general idea across). This is why MSAA falls down with a lot of different effects(alpha textures being the most common and obvious but also shader routines can introduce aliasing as is becoming increasingly obvious with modern games).

In the simplest terms- SSAA gives you a cleaner image; MSAA gives you a sharper image. While it isn't on a technical basis, the essence of SSAA is a blur filter. SSAA can help eliminate noticeable aliasing on anything, at the expense of high contrast levels. I fully understand why BFG likes SSAA, nothing is really comparable to how clean the image is using that method of AA. I personally am a big fan of razor sharp images which is why I am so obsessive about AF quality and why I like MSAA+AF better.

That makes perfect sense, and thank you for taking the time to elaborate. Is it possible that with NVIDIA's "new" SSAA, one can maintain their AF quality? It would seem that AMD's SSAA, giving a "cleaner," but more blurred image, goes hand-in-hand with it's rendition of AF (no aliasing, but some blurring), but NVIDIA would lose the benefit it seems to have over AMD in texture aliasing (or lack thereof) should their SSAA take off.

Oddly enough, those of us that do care already know what the differences are(thanks BFG ). This is why I stated those that don't know what the differences are probably aren't going to care- and for that I was attacked. How many times on this forum do people post things that help people learn anything? The reality is a bunch of flaming fanboys get all pissy if you don't champion their parts here, no matter if you are trying to teach them or not. This thread is yet another example of that. Yes, I do get an elitist attitude about it eventually as their are a small handful of people here that want to learn and a whole bunch that want to cheerlead their company of choice. You are currently a medical student- if we were in a thread talking about taking a certain prescription for a particular ailment I sure as hell wouldn't be bashing you for your take on why one was better then the other, I'd be asking questions to find out why(I understand that is what you are doing now, but look at the other posts in this thread).

Quite true, and I can see your problem and frustration when most threads here come down to fanboy bickering. I've found myself just jumping out of threads for the sake of it not being worth my time and effort to try to prove a point in all the mess that can accrue. However, anytime you wish to share some of your knowledge, I'll gladly listen. FWIW, I enjoyed reading through your thread on the "Architectural Direction of GPUs."

 

[GaiaHunter]

No, you can see the difference by simply testing and playing games. Once you get used to a certain level of image quality, it’s very hard to go back to anything worse because you spot the regressions immediately. Learning the theory just helps explain what you’re seeing.

Sure. I understand that.

Understanding or not the theory won't make it look better, for someone that just doesn't see the difference - although can make you appreciate the difficulties or the brilliantism of the implementation.

But while in those non-game pictures it is very clear the difference, in game, non-still some of those differences won't be so noticeable to everyone.

Specially if people just play newer games, that can't really use the higher IQ options.

I know many people play old games - I play mostly old games, as I keep finding older games to have better game play and/or story.

Anyway I just found that remark from Ben funny.

 

[GaiaHunter]

Some people don't 'believe' in global warming either.

That isn't a question of believing or not.

I guess no one says that climate isn't changing, just what causes it.

It is a question of people realizing our understanding about climate is far from comprehensive (look at weather forecasts, see if anyone can predict the weather in a certain area in a given day, or even a season, of some future year), which makes previsions based on computer models a laugh, since some parameters can be "tailored to taste", due to holes in our knowledge.

And then is of course the way we fight the climate changes. Reducing CO2 emissions is very very inefficient and expensive. There are cheaper ways to prepare to climate changes, that work regardless.

Especially because we have historical evidence that climate changes and has done so in the past, regardless of us, evil/greedy, human race being even on this planet.

But that is off topic and I apologize.

 

[scooterlibby]

26 pages?

I love the ceaseless quote by quote rebuttals.

 

[scooterlibby]

26 pages?

Yes, don't you now anything about the forums here? Threads aren't for staying on topic. Noob.

I love the ceaseless quote by quote rebuttals.

Which is exactly why you shouldn't be doing the same thing. Thank you for proving my point.

 

[Keysplayr]

Yes, don't you now anything about the forums here? Threads aren't for staying on topic. Noob.



Which is exactly why you shouldn't be doing the same thing. Thank you for proving my point.

Would the two of you like to be alone?

 

[Tempered81]

To the old pros: Does a negative LOD bias adjustment correct this so-called "blur" created by anisotropic filtering as it does correcting blur from sparse grid SSAA? Why do some AA modes fix jaggies on solid polygon edges, but do nothing to where Transparent textures touch polygon edges? Also, how can I fix shimmering with AA modes on a 4890? One thing I'm not too happy about is that i can't do SSAA on my 4890. Nor can I program the game config to render a larger pixeled image and downscale it into my screen to give that same "SSAA effect" which I believe is due to using VGA instead of DVI.

BTW, I've always favored ATI's AF & AA quality over Nvidia. Trubritar does a 5870 vs. 480 direct video comparison of Crysis benchmark, and you can clearly see where the Nvidia driver cheaps out and makes the game render distant mountain textures in a poor muddy quality compared to Ati. OTOH, I've seen a side-by-side image quality comparison in L4D & Crysis comparing Nvidia and ATi AF, and the Nvidia AF seems to produce sharper, better texture quality as the focal point becomes more distant (like ATi is stopping at 8x and Nvidia is going 16x and beyond).
Ati spent countless hours maximizing performance of crysis/warhead & 3dMark Vantage scores. Yet they lack that kind of optimization across the board. This is where Nvidia's comprehensive TWIMTBP program comes in and usurps Ati's few accomplishments. A 5870 might beat a 480 in crysis & vantage, but be 15%+ slower in 95% of the other games....

Nvidia's drivers are more performance related across the board through a wide gamut of PC games, which is a good thing to a certain extent until they begin to sacrifice quality. Ati always ends up trailing in performance on new releases (metro, bioshock, borderlands) After watching that Trubritar video, it's kinda lame to see the GTX480 producing muddy textures and coming in at the same speed as the HD5870. Nvidia's AA modes have made-up marketing names, and even sometimes the 16x and 32x filters look jaggier than a 2x SSAA on Radeons.

Disregarding price, I would take a 480 or 470 over a 5800 card. When you figure price into the equation Ati looks better. I got a sweet deal on a used 4890 - about half the price of comparable performance in GTX200 series cards. Honestly I would love to have a 480 SSC or a Toxic 5970... heheh that is a ton of cash though...

Generally speaking, the Nvidia drivers play better with games across the board.

 

[BFG10K]

I took a quick read through, and I see where the GTX 285's rendering gave a sharper texture. I would ask then, is this only in OpenGL-based games or is it universal? Also, has the 5xxx series AF been tweaked in drivers or is it a limitation of the architecture?

It’s universal and all signs point to it being an architectural decision. The 5000 series is angle invariant but it also undersamples compared to nVidia’s parts. This causes visible transitions during in-game movement and also reduced sharpness compared to nVidia’s parts. The tester app also showed more texture aliasing at closer distances.

My other inquiry would be to see if you think your GTX 470 has an even better rendering method than the GTX285. You said you prefer the GTX 285's method of AF (minimal aliasing with sharper textures) as opposed to AMD's method of AF (no aliasing with more blurry textures), but what has Fermi brought to the table in this regard?

The AF on the GF100 is the same as it on the G8x/G9x/GT2xx, but there have been notable changes to AA. I’ll be doing a full IQ piece on the GF100 soon, and I’m already constructing a mental image of what I want to cover.

That makes perfect sense, and thank you for taking the time to elaborate. Is it possible that with NVIDIA's "new" SSAA, one can maintain their AF quality?

AF and SSAA are two completely different things. They can affect each other, but they’re a completely different class of image enhancement. nVidia’s hidden SSAA is exactly the same as ATi’s (and it also allows decoupled samples) so when both are in operation with AF, nVidia’s parts still look better.

 

[BFG10K]

It is interesting that AnandTech shows completely different findings from what is reported by BFG and Ben.

Not being able see something doesn’t prove it isn’t there. Look my images – the evidence is right here.

Image Quality
1. Look at the vegetation at the very top right of the screen.
4890 DX9: 4AA + AAA (adaptive AA) >>> GTX480 32x or even GTX285 DX9: 4AA + TrSS

2. Look at the vegetation right above the gun and the palm tree leaves, as well as the longer end for the leaves being shaded in the top right hand corner.
5870's DX9: 4AA + SSAA >>> GTX480 DX9: 4AA + TrSS.

On the 4000 series AAA was inferior to TrAA. nVidia did more work and had a more accurate image, while ATi tended to fatten alpha textures. Also AAA’d textures would shimmer more than TrAA’d textures during in-game movement because the weaker AF caused more texture aliasing.

It’s quite possible the 5000 series has improved AAA but again, a weaker AF would counter this. This is also something that has to been seen during in-game movement to really see what is happening.

Stating that GTX285 does AF better than GTX470 is misleading.

I don’t think anyone stated that because the GF100 is the same as the G8x/G9x/GT2xx with regards to AF.

"More to the point, we have yet to find a game where the difference between AMD and NVIDIA’s AF modes have been noticeable; so technically AMD’s AF modes are better [With the Radeon 5870 we saw AMD implement true angle-independent AF]"

Anyone used to gaming on nVidia’s parts will spot the regression immediately during in-game movement. The 4000 series shimmers far more, while the 5000 series has visible filter transitions.

My screenshots show the differences and the effects become more apparent during in-game movement.

 

[BFG10K]

To the old pros: Does a negative LOD bias adjustment correct this so-called "blur" created by anisotropic filtering as it does correcting blur from sparse grid SSAA?

No, and you should never use a LOD bias adjustment to correct AF. If you have to do this it means the AF is broken. The whole purpose of AF is to sharpen the image without additional aliasing. A negative LOD bias will introduce more aliasing, and will do nothing for the filtering transitions either.

The only time you should be using a negative LOD bias is with SSAA not generated from oversampling, which typically means to RG and SG variants.

Why do some AA modes fix jaggies on solid polygon edges, but do nothing to where Transparent textures touch polygon edges?

MSAA anti-aliases based on Z (depth) values; transparent textures have no depth values since they’re not geometry, so it doesn’t work on them. To affect them you need to use SSAA (either full scene or localized TrAA/AAA) because SSAA retains full texture and shader samples.

BTW, I've always favored ATI's AF & AA quality over Nvidia. Trubritar does a 5870 vs. 480 direct video comparison of Crysis benchmark, and you can clearly see where the Nvidia driver cheaps out and makes the game render distant mountain textures in a poor muddy quality compared to Ati.

I can’t comment on this, but I co-wrote a Crysis piece a while ago with G8x vs 4000, and there was no evidence of this happening. In fact, certain combinations of settings caused lower image quality in-game on ATi’s parts.

Nvidia's AA modes have made-up marketing names, and even sometimes the 16x and 32x filters look jaggier than a 2x SSAA on Radeons.

This is nothing to do with ATi vs nVidia but rather with the class of AA being used. SSAA will almost always look better than MSAA while taking less samples because it affects everything in the scene, unlike MSAA which only affects polygon edges. That and the effects of edge-only AA rapidly diminish after 4xAA unless you start super-sampling. This is why it would be a travesty for nVidia to remove the hidden SSAA modes.

 

[evolucion8]

MSAA takes geometric data and creates a higher fidelity "image" of Z values and then compares them locating pixel areas that have varriations in Z value and then samples multiple color samples only from those pixels(in a nutshell, a bit more complicated then that but that gets the general idea across). This is why MSAA falls down with a lot of different effects(alpha textures being the most common and obvious but also shader routines can introduce aliasing as is becoming increasingly obvious with modern games).

It could be easier if instead of copy/pasting articles from the internet, just said that MSAA simply samples the Z-Buffer, period.

 

[BenSkywalker]

It could be easier if instead of copy/pasting articles from the internet, just said that MSAA simply samples the Z-Buffer, period.

You think I copied that from somewhere? Link it up or shut your mouth.

Edit

I was going to leave it at that but I will get a bit more in depth on the subject matter. In the build up to the release of 3dfx's VSA100 based parts they released a white paper on exactly how they were handling AA on their parts. After reading their white paper I explained to them particular issues that they were going to have with improper LOD clamps due to the method they were utilizing and how it was going to result in excessive blurring. Two of them registered on this forum as a result of that, Dave Barron and Kristof(can't recall his last name, Dave Barron is easy as we had to keep him and Dave Baumann straight- Kristof was working for PowerVR last I was aware, Dave went to work for BitBoys- not sure what he is up to now). Both of them insisted I was wrong on the issue- the parts shipped and everyone complained how it blurred images to much and they issued a fix to adjust LOD bias to properly compensate for their rendering techniques. I have a rather extensive background in dealing with AA implementations since years prior to them being available in real time.

 

[SirPauly]

Kristof Beets.

 

[Tempered81]

Thx for the replies, BFG. And here is that video I was talking about Nvidia drivers cheaping out on distant textures:

http://www.youtube.com/watch?v=SlEA8E9NomM

-check between time 1:15 and 1:25.

Check out the difference in texture quality on that mountain face. Obviously a driver optimization shortcut they hoped would go unnoticed.

 

[dust]

Yeah, I've seen that in trubitar's bench, it seems pretty clear the gtx image is actually worse and the driver shortcuts you are referring to didn't help getting higher performance also. On a side note I noticed the same story with a XFX 285 vs my 4870's, one of the main reasons I haven't exchanged them for it.
Maybe BFG is too close to the project ;-D, but, I guess the whole thing is rather subjective.

 

[evolucion8]

You think I copied that from somewhere? Link it up or shut your mouth.

Yeah, you are very brave and rude behind your keyboard, you are the MAN! I want to be like you!! (NOT), but I don't think that you will dare to say that on my face, dummy :awe:

 

[BenSkywalker]

Yeah, you are very brave and rude behind your keyboard

You accuse me of plagiarism behind your keyboard, back it up or shut your mouth.

I don't think that you will dare to say that on my face

You some sort of internet tough guy? Hehe, of course in real life you are a seven foot tall 350lb MMA master that everyone should fear, of course you are.

 

[v8envy]

You some sort of internet tough guy? Hehe, of course in real life you are a seven foot tall 350lb MMA master that everyone should fear, of course you are.

For some reason I have no trouble buying into any forum warrior claiming to be 350 pounds. The other stuff, not so much.

 

[evolucion8]

You accuse me of plagiarism behind your keyboard, back it up or shut your mouth.

You are the one who think that you are the tough internet guy, telling people to shut their mouth, who the hell you think you are? Standing behind a keyboard defying people, that's funny dummy :awe:. I'm certainly sure that you wouldn't say it in my face, sure for that. So va a' forti fottere.

I may not weight 350lbs or being 7f tall, but I'm not fat like you