Mod, please lock, this is no longer on topic.

[BFG10K]

The R100 didn't actually offer 64 tap anisotropic filtering.

Yes it did and your opinion doesn't change fact. Granted the implementation was sub-par and cut a lot of corners but that still doesn't change fact.

I want to use the best available method of reproduction

So we're talking about pre-rendering only which means no more gaming for you since games are mathematically inferior to pre-rendering and you'd have to be d@mn blind to play them. I'm glad we sorted that out.

I consider considerably increased aliasing and misaligned LOD selection on adjacent textures very pronounced IQ reduction.

Now you're jumping between the original R100 and modern cards whenever it suits you. Modern cards have neither considerably increased aliasing or misaligned LOD. You're blowing this way out of proportion and making it sound like 320x240 software point filtering or something, then in the same breath you're more than happy to drop MSAA and AF or you're more than willing to run slideshows at low/middling resolutions.

Depends on the particular game- Pacific Fighters is showing ~10% performance hit.

Wow, a whole 10%. Let me a get a chair before I fall over from the shock. :roll:

But 16xAA running Doom III on a SLI setup at 26 FPS @ 1600x1200, now that's viable according to you.

Haven't had enough hands on time with it to make a call there- but it obviously is much better then the poor hack of MSAA.

"Poor hack". So again you'd rather go without? "I want the best possible reproduction available but I won't run MSAA or AF because it's not real" or "I'll run 16xAA at 1600X 1200 @ 26FPS".

Remind me again, which of those scenarios offers "the best possible reproduction?"

BS.

Got any links to back that up? More specifically, running "real AF" with that too?

Using that logic none of the current parts can run 256x AF so obviously they can't run AF at all.

Exactly, according to your own reasoning. So what's your response to that then? I claim 8xAF/16xAF isn't "real" but 256xAF is because it's mathematically provable, using your own logic. I guess this means you can't even use the NV2x's AF either, huh?

If it changes the ouputted pixel yes. I want to see it changes the outputted pixel or not.

Changes it from what?

Trunctuating non visible data isn't an issue.

Visible? So now it's what's visible? So what happened to your mathematically provable BS?

And who decides if it's visible? You? Me? Are you now telling us that the definition of "real" AF is what is visible and what isn't?

I think I can see just what you mean by "real": "real" is what I (Ben) says it is whenever it suits me.

 

[BenSkywalker]

So we're talking about pre-rendering only which means no more gaming for you since games are mathematically inferior to pre-rendering and you'd have to be d@mn blind to play them. I'm glad we sorted that out.

Reproduction of graphic elements designed to be rendered in real time would be reproduction of graphics in real time. Sorry to confuse you by speaking with exacting language- I will address this again later in the post.

Modern cards have neither considerably increased aliasing or misaligned LOD.

Yes, they most certainly do.

You're blowing this way out of proportion and making it sound like 320x240 software point filtering or something, then in the same breath you're more than happy to drop MSAA and AF or you're more than willing to run slideshows at low/middling resolutions.

I'm not the one that keeps brining up 320x240 or software filtering- that is you. Is it due to current rasterizers reminding you of it so much? I will say again the Final Fantasy the Spirits Within looks astoundingly better running 720x480 then anything I have seen to be rendered by a consumer graphics card in real time at any resolution or settings.

But 16xAA running Doom III on a SLI setup at 26 FPS @ 1600x1200, now that's viable according to you.

Am I speaking too clearly for you to comprehend? I stated that SLId GTXs have enough power to handle 4x SSAA- that is FOUR. What sort of twisted dialect should I be using?

"Poor hack". So again you'd rather go without? "I want the best possible reproduction available but I won't run MSAA or AF because it's not real" or "I'll run 16xAA at 1600X 1200 @ 26FPS".

Remind me again, which of those scenarios offers "the best possible reproduction?"

Are you capable of basic reading comprehension? I'm having doubts. I'm no longer going to try and answer questions you create in your mind that are completely removed from what I have said.

Got any links to back that up? More specifically, running "real AF" with that too?

Scroll down for 4xS on a 5900. Running 1600x1200 the 5900Ultra takes a 60% performance hit running 4xS versus no AA, versus 31% for 4x MSAA. That is the last part 4xS was exposed on- and even though it was nV's weakest core(NV3X) in relative terms that they have produced since the TNT it was still far from a staggering performance difference in absolute terms. Apply the additional performance hit to newer titles.

Exactly, according to your own reasoning. So what's your response to that then? I claim 8xAF/16xAF isn't "real" but 256xAF is because it's mathematically provable, using your own logic. I guess this means you can't even use the NV2x's AF either, huh?

Show me a part that supports genuine 256xAF and I'll buy it. Again, my apologies for saying exactly what I mean when using terms like available.

Visible? So now it's what's visible? So what happened to your mathematically provable BS?

Were you particularly weak in mathematics? When trying to analyze if something is proper or proveable you need look to the answer- not the equation. The answer for the topic we are discussing currently is the outputted pixel values. If those values are correct- then you can not on a mathematical basis prove them to be otherwise. It really is that simple. As I stated back in your days of bashing nVidia in so many threads for 'cheating'- if it shows up- rip them. If the outputted pixel is exactly the same then it doesn't matter.

And who decides if it's visible? You? Me? Are you now telling us that the definition of "real" AF is what is visible and what isn't?

We got off on this tangent as I was asking you which drivers introduced shimmering on the NV2X cores- a question you never did answer that I saw. What determines if it is visible is what is drawn on the screen. I know you aren't going to see much of anything- I was dumbfounded at how extremely poor the R3x0 parts were at AF after hearing you speak so highly of it(particularly performance mode- yech). Back to back or side by side the increase in aliasing was shocking. Now I incorrectly assumed for a long time that this was due entirely to poor LOD selection- it turns out it was due to flat out inferior filtering hardware(reduced accuracy in blending ops significantly reducing IQ).

I think I can see just what you mean by "real": "real" is what I (Ben) says it is whenever it suits me.

Point me out to when I have used a double standard. Feel free. If there are rendering errors or issues then I will gladly bash whoever it is at fault and I was doing this years before you registered here(a while before FuseTalk was around actually).

 

[ArchAngel777]

There is no doubt about it, Ben you are an extremist to the point that your opinion matters little. Only your eyes can see these "artifacts" and only you would care so much about them to discuss them constantly. Is 16X Performance AF image quality worth the 10% performance loss? I do, and I do not think that can be disputed. I agree with BFG that it does improve image quality and perhaps not to the best, which is what I argued against in the first place, it does offer decent filtering for the performance hit.

I am not sure anyone can take you seriously on this forum. Your statements are extreme hyperboles. If things were really as bad as you say, then I think we would see more articles on this type of stuff. In fact, you would probably see Anand himself talking about it.

Seriously, you need to see a shrink or something. Because the little spec or dust on your monitor would drive you insane. How can you live? Certainly not healthy... Do you complain about everything in your life? Your table? Your chairs? Your bed? Your car? Your house?

Plain and simple is that games have never looked better in all of PC gaming and they will continue to look better and better. If the result is a great looking game, I could care less what filtering is used. Hell, if you can make a game look better than 16X HL2 and that game does not use any AF or AA, who the hell cares? Your end result is a better looking game.

Go on, keep on with your hyperboles... As someone who has read all the posts here, you started out strong, but now your position is weak and full of bias. Come back when you can learn how to post more reasonably.


EDIT ** I know you cannot have any joy in your life. I mean, your statement that MP3's sound horrible and you cannot stand to listen to them. LOL, go for it... The difference is, I enjoy the music and you do not. You will sit there and listen to anything that does not sound right. That is the difference between you are the rest of the world. You *look* for the bad rather than enjoying the good. It seriously must suck to live a life that way. Enjoy life a little and quit looking for a problems. Trust me, you will enjoy life.

You have a mindset that looks for problems. In fact, you are so obsessed with problems that it blocks out your ability to see what is right and true... Hey, it is your life, just trying to give you some pointers. Because there is no way in hell you can sleep at night without finding something wrong...

 

[xtknight]

Originally posted by: BenSkywalker

Modern cards have neither considerably increased aliasing or misaligned LOD.

Yes, they most certainly do.

Question: How does a card increase aliasing? You're talking about the decreased AA quality?

 

[ArchAngel777]

Originally posted by: xtknight

Originally posted by: BenSkywalker

Modern cards have neither considerably increased aliasing or misaligned LOD.

Yes, they most certainly do.

Question: How does a card increase aliasing? You're talking about the decreased AA quality?

In addition to this question, you have talked a lot about "math" but nothing to back up those claims. I am not saying you do not have the proof, but I am saying - Show it to me.

I would like to know the math behind your correct AF versus performance AF. I will await your formula's and hope they are well versed. To clarify, I am asking you how you know the correct mathematical formula for correct AF and the current one for adapative AF. I would some numbers and accurate information. Since this is a forum, you should have ample time to look it up on google to build your case.

Show me the substance.

 

[BFG10K]

Reproduction of graphic elements designed to be rendered in real time would be reproduction of graphics in real time.

Err, okay, let's try something different then. Since Age of Empires 3 is mathematically provable to be superior to Sacrifice and offers the best possible reproduction for a RTS at this time, you will never play Sacrifice again and therefore we will never hear you lambast ATi for the W-buffer issue again.

I can't wait to see you spin yourself out of that one.

Yes, they most certainly do.

No they don't.

I will say again the Final Fantasy the Spirits Within looks astoundingly better running 720x480 then anything I have seen to be rendered by a consumer graphics card in real time at any resolution or settings.

What the hell does have to do with the topic at hand?

I stated that SLId GTXs have enough power to handle 4x SSAA- that is FOUR. What sort of twisted dialect should I be using?

I'm sorry, but when did 4xS become "the best possible reproduction available"? Another double-standard I see?

Given that plain 4xSSAA is inferior to 6xMSAA in 90% of cases I really was expecting you meant 16xAA which encapsulates 4xSSAA. I gave you the benefit of the doubt but it looks like that benefit was was misplaced. It's just another case of your "I'll only use the best possible reproduction" go out the window yet again because it happens to suit you at this time.

Are you capable of basic reading comprehension?

Are you? You keep producing examples that you allege are viable on a 7800 SLI setup so tell me Ben, what were you doing when such a configuration wasn't available? Either you were not using MSAA or you were running at slideshow speeds, neither of which offer "the best possible reproduction". In otherwords you were running inferior IQ to the masses.

Not only that but AFAIK you don't even own a 7800 SLI setup so really you appear to be arguing with thin air using nothing but hypothetical scenarios and guesswork to back your claims.

Scroll down for 4xS on a 5900. Running 1600x1200 the 5900Ultra takes a 60% performance hit running 4xS versus no AA,

60%? Yet you were worried about one game where you found 10% with 16xAF. This proves my exact point about how unbalanced and inconsistent your point of view is.

Not only that but Quake 3 is CPU bound so if you look at the UT2003 flyby benchmarks 4xS causes the card to run at 39% of its original speed @ 1600x1200. Add to the fact that AF is disabled in the 4xS list and on top of that "real" AF would cause a much bigger hit than than optimized AF and you'd probably be looking at 5% of the original performance with"real" 8xAF and "real" 4xS, at a glorious middling resolution to boot.

This example perfectly illustrates the ludicrous reasoning you're employing.

Again, my apologies for saying exactly what I mean when using terms like available.

Which goes back to my original question which was never answered: where are you using that Ti4600's "real" 8xAF of yours Ben? Please, list the games, the settings and their performance in said situations. In particular I'm interested in newer titles such as Doom 3, Far Cry, Riddick and BF2.

When trying to analyze if something is proper or proveable you need look to the answer- not the equation. The answer for the topic we are discussing currently is the outputted pixel values.

Yup, and outputted pixel values using 256xAF and 32 bit filtering are mathemetically provable to be superior to 8xAF and 8 bit filtering. Oh that's right, my example isn't available so mathematics go out the window for that one. But AOE3 is available and is also mathematically provable to be superior to Sacrifice or any other RTS so I expect you will never touch another RTS again.

We got off on this tangent as I was asking you which drivers introduced shimmering on the NV2X cores- a question you never did answer that I saw

Any driver that allows the functionality of reducing the filtering on texture stages.

I know you aren't going to see much of anything-

Sorry, are we talking about your selective nVidia-vision now?

Point me out to when I have used a double standard

Done repeatedly.

 

[BFG10K]

Question: How does a card increase aliasing? You're talking about the decreased AA quality?

No, he's talking about reduced samples/precision in AF. Of course he's blowing the net effect way out of proportion.

 

[BenSkywalker]

Since Age of Empires 3 is mathematically provable to be superior to Sacrifice

Explain how it is proveable. In my examples the end result was a pixel output which is the entirety of the purpose of what we are discussing. Given differing factors you could state that CivIII is vastly superior to AoE3, or RoN to both of them, or Sac to all of them, and even then you could make an argument for StarCraft- it depends on what factors are weighed. In terms of dealing entirely with a mathematical equation- which is all graphics hardware does- giving the correct answer is what matters.

No they don't.

I will make note of the fact that you are also the person that was raving about bilinear filtering on your R9700Pro.

What the hell does have to do with the topic at hand?

Resolution helps make graphics sharper- I would gladly give up resolution to jump to the level of rendering used in FFTSW- you wouldn't?

I'm sorry, but when did 4xS become "the best possible reproduction available"? Another double-standard I see?

I have been saying the exact same thing in terms of AA since the beginning of this conversation- why is it just now a double standard? You created the other standard in your head.

Given that plain 4xSSAA is inferior to 6xMSAA in 90% of cases

You have that backwards. MSAA does nothing at all for the overwhelming majority of the scene.

I really was expecting you meant 16xAA which encapsulates 4xSSAA. I gave you the benefit of the doubt but it looks like that benefit was was misplaced.

I can't speak of 16xAA on consumer parts in real time as I haven't seen it in action yet. I certainly don't trust others eyes.

You keep producing examples that you allege are viable on a 7800 SLI setup so tell me Ben, what were you doing when such a configuration wasn't available?

I'm saying the performance levels would make it viable- I'm not saying it is the ideal setup as what I have been stating isn't even possible on a 7800GTX SLI setup as of now. Why are you thinking I'm saying anything else? Nowhere did I give that indication.

Not only that but AFAIK you don't even own a 7800 SLI setup so really you appear to be arguing with thin air using nothing but hypothetical scenarios and guesswork to back your claims.

Why would I purchase parts that I know I can't stand the rendering of? Have you not been listening to anything I have been saying? You are trying to jump to conclusions on what I am meaning instead of listening to what I am saying- that won't get you anywhere.

if you look at the UT2003 flyby benchmarks 4xS causes the card to run at 39% of its original speed

READ-

Running 1600x1200 the 5900Ultra takes a 60% performance hit

You are pointing out not only what I stated, but what you quoted me as saying. I WAS talking about the UT2K3 numbers.

Add to the fact that AF is disabled in the 4xS list and on top of that "real" AF would cause a much bigger hit than than optimized AF and you'd probably be looking at 5% of the original performance with"real" 8xAF and "real" 4xS, at a glorious middling resolution to boot.

34% of the original speed. That is with real FSAA and real AF. Obviously an enormous performance hit, but if you are largely CPU bound anyway as you would be at the performance levels of the 7800GTX then you can afford to drop some frames. I could have linked to numerous examples showing significantly smaller hits btw(Dungeon Siege, UltimaIX)- I looked for one that was GPU limited. The performance hit of SSAA and real AF is not cumulative- you are straining differing elements of the GPU in the process of running both at once.

Which goes back to my original question which was never answered: where are you using that Ti4600's "real" 8xAF of yours Ben?

To quote myself yet again-

I want to use the best available method of reproduction

Now considering that the Ti4600 isn't even capable of playing some newer games it is obviously out of the question there- not to mention that it is not a singular rendering issue. That said- I am more then willing to take the performance hit to render it properly on newer hardware if they would simply offer the choice. The Ti4600 is not an available solution to render anything if the game won't play at all on it.

Yup, and outputted pixel values using 256xAF and 32 bit filtering are mathemetically provable to be superior to 8xAF and 8 bit filtering.

Absolutely.

Oh that's right, my example isn't available so mathematics go out the window for that one.

Not in the least- show me where I can buy a part that can handle it.

But AOE3 is available and is also mathematically provable to be superior to Sacrifice or any other RTS so I expect you will never touch another RTS again.

You are confusing subjectivity with science. You really shouldn't.

Any driver that allows the functionality of reducing the filtering on texture stages.

Which brings us full circle- WHICH DRIVERS??? I will install them and check them out- but I need to know which ones you are talking about.

Sorry, are we talking about your selective nVidia-vision now?

Last time I purchased any nVidia graphics card- January of 2000. That was seven months before you registered on this forum. In all the years we have been discussing hardware I have not purchased a single nVidia graphics card for use in any of my machines. Serious nV vision huh?

Done repeatedly.

Where? You have speculated and misread several things- show me one singular example of a double standard.

Of course he's blowing the net effect way out of proportion.

But you do not deny it exists.

I would like to know the math behind your correct AF versus performance AF. I will await your formula's and hope they are well versed. To clarify, I am asking you how you know the correct mathematical formula for correct AF and the current one for adapative AF.

The correct version of AF is sampling is non square. That is by definition of the word. Check out the filtering/color wheels numerous sites post and compare current parts to one of the boards that does it properly- they aren't close.

 

[ArchAngel777]

The correct version of AF is sampling is non square. That is by definition of the word. Check out the filtering/color wheels numerous sites post and compare current parts to one of the boards that does it properly- they aren't close.

Could you be any more vague? I am asking you to provide me with the proper mathematical formula behind this "correct AF". So to speak... Pretty much all I am interested in. I just want to know this correct procedure and how we know it is the only to get accurate filtering.

 

[BenSkywalker]

Could you be any more vague?

Anisotropic filtering has a definition- non square filtering. There isn't an IEEE standard associated with it. In order to properly perform anisotropic filtering the sampling must be non square by definition. I mention the color wheels as you will see that newer parts regularly fail to maintain this standard across the scene. They are by definition of the word not doing it properly.

I just want to know this correct procedure and how we know it is the only to get accurate filtering.

It would be possible to approach it from several different angles and have it done properly. As an example there are Talisman and Texram techniques(I'm partial to Talisman as samples are weighted giving better IQ IMO- although that is subjective). Both of those work their filtering by using a quadrilateral principle of sampling and adjusting the sampling methodology according to the relative perspective of the scene- samples are blended evenly in Texram or weighted using Talisman. These are a couple of simple examples- there are numerous ways you could approach AF and stay within the definition of the word- at that point it would become a subjective issue. None of the current hardware remains inside the definition as they fail to sample in a non square pattern across the entire scene(they aren't even close actually).

 

[kylebisme]

Originally posted by: BenSkywalker

Could you be any more vague?

Anisotropic filtering has a definition- non square filtering. There isn't an IEEE standard associated with it. In order to properly perform anisotropic filtering the sampling must be non square by definition. I mention the color wheels as you will see that newer parts regularly fail to maintain this standard across the scene. They are by definition of the word not doing it properly.

I just want to see if I am following your logic here. First off, anisotropioc means non-square and the highest quality filtering presents a rounded appearance in certian anisostripic filtering tests (which I assume is what you are referring to as "color wheels"); so the other lesser quality methods are not really anisotropy because they look pointy. Is that how your argument flows?

 

[BenSkywalker]

First off, anisotropioc means non-square and the highest quality filtering presents a rounded appearance in certian anisostripic filtering tests (which I assume is what you are referring to as "color wheels"); so the other lesser quality methods are not really anisotropy because they look pointy. Is that how your argument flows?

Look up how LOD bias selection works on various hardware- talk to the driver teams about how it is that a LOD bias selection is chosen and ask them if this is fixed or flexible depending on the samples taken and then apply that information to the color wheels. You'll figure it out pretty quick.

If there was some sort of major flaw with my simplification of using that BFG would certainly have pointed it out by now- he likes to point out any potential flaw in my arguments. The issue we are debating is how much of an impact this has- BFG considers it minimal while I consider it horrible.

 

[BFG10K]

Explain how it is proveable.

Quite easily. Simply using HDR, it's mathematically provable that it matches the properties of real light closer to standard light rendering.

Of course you'd never admit this because then you'd be forced to apply your ridiculous logic to Sacrifice (no more Sacrifice for you since AOE3 is now the "best possible rendering") and it would expose the nature of your whole argument. This is why you continue to weasel out of each refutation I produce for your flawed arguments.

I will make note of the fact that you are also the person that was raving about bilinear filtering on your R9700Pro.

Yup, and it looked a thousand times better than a Ti4600's straight trilinear and almost as good as the Ti4600's trilinear AF. You can quote me on that anytime you like.

have been saying the exact same thing in terms of AA since the beginning of this conversation- why is it just now a double standard? You created the other standard in your head.

What the hell are you talking about?

You wanted "best possible reproduction" (and that's a direct quote). 4xSSAA - the flawed example you introduced in order to save face that SSAA is generally unusable - is not the best possible reproduction. 16xAA is.

So Ben, why aren't you lambasting 4xSSAA that you'd have to be d@mn blind to use it and advocating 16xAA instead? Not only are you not attacking it, you actually produced 4xSSAA as a scenario to back your own arguments.

Seriously, you should quit while you're ahead. Your arguments have so many holes in them they're like a collection of cheeses in a Swiss shop.

You have that backwards. MSAA does nothing at all for the overwhelming majority of the scene.

What else is there? Polygon internals (which is taken care of by AF) and alpha textures (which are far and few between). It's a pointless waste of resources to blindly AA everything, especially now thanks to nVidia's TAA. Of course that's not real AA by your definition so you won't be using it either? I guess you're SOL, huh? My, you must have a difficult life indeed.

I can't speak of 16xAA on consumer parts in real time as I haven't seen it in action yet.

So it's the best possible AA available yet by your own admission you haven't even seen it in action? That doesn't really leave much credibility in your claims, does it?

I'm saying the performance levels would make it viable-

But they don't make it viable and this is exactly why you've backpedaled from "to best possible reproduction" to 2048x1536 + 0xAA or 1600x1200 + 4xSAA. Best possible reproduction is 2048x1536 with 16xAA. Like I said before, your arguments are obtuse to begin with but when you backpedal like this your credibility goes out the window.

Why would I purchase parts that I know I can't stand the rendering of?

Because 16xAA is the best possible AA available and to quote yourself "I look for the best possible reproduction".

34% of the original speed. That is with real FSAA and real AF. Obviously an enormous performance hit,

Yup and not only that on the previous page their resolutions are 1152x864 and 1280x1024. So if you class 1920x1440 as a low resolution how can anyone take you seriously with these results when you present yet another blatant double-standard?

but if you are largely CPU bound anyway as you would be at the performance levels of the 7800GTX

But you wouldn't be CPU bound, especially not at "best possible reproduction levels". Again that means about 26 FPS in Doom 3 and that's still at a low resolution of 1600x1200 (according to your standards) with optimized AF and 16xAA. Real AF @ 2048x1556 and 16xAA? You'd be looking at 5 FPS if you're lucky on a 7800 SLI setup. Yup, real viable there Ben. :roll:

I could have linked to numerous examples showing significantly smaller hits btw(Dungeon Siege, UltimaIX)-

Uh huh, and which of those examples were running at "best possible reproduction?" Where are those 2048x1536 bechmarks Ben?

The performance hit of SSAA and real AF is not cumulative- you are straining differing elements of the GPU in the process of running both at once.

Now that's just nonsense and you know it. While the performance characteristics aren't exactly the same in the two examples, fundamentally SSAA and AF both stress memory bandwidth and texel fillrate. I mean your own results back the cumulative performance hit for heaven's sake.

Which brings us full circle- WHICH DRIVERS??? I will install them and check them out- but I need to know which ones you are talking about.

After the antics you've performed here, if I go down this tangent do you really expect me to believe that you'll admit to shimmering? Or that if you even find shimmering that you won't find some way to weasel out of it?

Last time I purchased any nVidia graphics card- January of 2000. That was seven months before you registered on this forum. In all the years we have been discussing hardware I have not purchased a single nVidia graphics card for use in any of my machines.

And that means what exactly?

But you do not deny it exists.

I never once said opmitized AF is an exact match to real AF. What I said was the IQ loss is minimal compared to the performance gain you get.

Check out the filtering/color wheels numerous sites post and compare current parts to one of the boards that does it properly- they aren't close.

Even the R100 was employing non-square sampling patterns in the direction of the stretch. Why don't you go and read any whitepaper on the subject instead of continuing with your fictional interpretations?

None of the current hardware remains inside the definition as they fail to sample in a non square pattern across the entire scene(they aren't even close actually).

Whether you need a non-square sampling pattern depends entirely on each surface in the scene. A good AF algorithm shouldn't apply AF to surfaces that don't need it.

 

[BenSkywalker]

Quite easily. Simply using HDR, it's mathematically provable that it matches the properties of real light closer to standard light rendering.

You are talking about the graphics engine and ignoring gameplay. Gameplay is why I buy games- I buy hardware for rendering. I have no clue how your are equating gameplay with graphics.

Yup, and it looked a thousand times better than a Ti4600's straight trilinear and almost as good as the Ti4600's trilinear AF.

Aliasing and mip banding included.

You wanted "best possible reproduction" (and that's a direct quote). 4xSSAA - the flawed example you introduced in order to save face that SSAA is generally unusable - is not the best possible reproduction. 16xAA is.

So Ben, why aren't you lambasting 4xSSAA that you'd have to be d@mn blind to use it and advocating 16xAA instead? Not only are you not attacking it, you actually produced 4xSSAA as a scenario to back your own arguments.

Seriously, you should quit while you're ahead. Your arguments have so many holes in them they're like a collection of cheeses in a Swiss shop.

Show me a part that has the power to make it playable. Current graphics hardware(anything nV for some time and likely the R520) are capable or handling off line rendering which would be vastly superior to any game we will see in years. That could take hours or days per frame though. Where have I suggested that I want rendering at sub playable speeds? What I have been saying that SLId GTXs have the power to handle it. Have you seen me advocate a music device that could only reproduce the song with stuttering? Or movies for that matter? Obviously if it can't muster playable speeds it is not reproducing the gameplay experience the developers intended.

What else is there? Polygon internals (which is taken care of by AF)

Only on angles so deemed worthy by nV and ATi- certainly not all of the internals.

and alpha textures (which are far and few between).

All over the place in HL2 and lots of them used in FarCry too.

It's a pointless waste of resources to blindly AA everything, especially now thanks to nVidia's TAA. Of course that's not real AA by your definition so you won't be using it either?

It is real AA as is MSAA- it is not real FSAA. Mip mapping is real AA. I don't have enough hands on time with it to judge it- I haven't seen if it fails under certain conditions as current AF schemes do a great deal of the time.

So it's the best possible AA available yet by your own admission you haven't even seen it in action? That doesn't really leave much credibility in your claims, does it?

The only parts offering it have very poor texture filtering at best. Given that texture aliasing covers vastly more screen real estate I don't see how you couldn't possibly understand why that negates any miniscule advantage that anything north of 4x AA @20x15 would provide.

But they don't make it viable and this is exactly why you've backpedaled from "to best possible reproduction" to 2048x1536 + 0xAA or 1600x1200 + 4xSAA. Best possible reproduction is 2048x1536 with 16xAA. Like I said before, your arguments are obtuse to begin with but when you backpedal like this your credibility goes out the window.

What am I backpedaling from exactly? You are trying to make it out like I said something I never have come remotely close to hinting. I never once said anything like it is OK if the game is completely unplayable in fact that is why I am suggesting the $1K plus setup as the option where solid rendering would be viable on today's games. You have created this non playable scenario in your mind and insist that I am the one that came up with it- it is ALL you.

Because 16xAA is the best possible AA available and to quote yourself "I look for the best possible reproduction".

Not remotely close to the best, I was rendering using higher sampling then that with stochastic sampling patterns ten years ago.

Yup and not only that on the previous page their resolutions are 1152x864 and 1280x1024. So if you class 1920x1440 as a low resolution how can anyone take you seriously with these results when you present yet another blatant double-standard?

You find numbers with 4xS in use at high res- I can't. I've looked, I posted the highest settings I could find. The point of posting those was your insinuation that the performance hit for SSAA and AF was cumulative which it isn't.

But you wouldn't be CPU bound, especially not at "best possible reproduction levels".

Obviously I was clearly indicating when you WERE CPU bound you had overhead to spare. I didn't think that needed much more clarification.

Again that means about 26 FPS in Doom 3 and that's still at a low resolution of 1600x1200 (according to your standards) with optimized AF and 16xAA. Real AF @ 2048x1556 and 16xAA? You'd be looking at 5 FPS if you're lucky on a 7800 SLI setup. Yup, real viable there Ben.

And what exactly is your point?

Running 1600x1200 in D3 without AA(and with hack job AF) the 7800GTX is still largely CPU limited and pushing just under 90FPS. Say it is entirely GPU limited at that setting and take the performance hit the Ti4200 had from real AF and 4xS and it would be pushing roughly 30FPS(that is ignoring any hit due to D3's built in useage of hackjob AF). That's running 1600x1200 in the most visually impressive game to date. That is obviously much lower res then ideal, to get decent playability I would likely drop to 2xS and up the resolution if we were talking GTX levels of performance.

Uh huh, and which of those examples were running at "best possible reproduction?" Where are those 2048x1536 bechmarks Ben?

They weren't there BFG. Why don't you link to them since I was obviously trying to hide something. Since 2048x1536 testing was so incredibly commonplace prior to the launch of the GTX.

Now that's just nonsense and you know it. While the performance characteristics aren't exactly the same in the two examples, fundamentally SSAA and AF both stress memory bandwidth and texel fillrate. I mean your own results back the cumulative performance hit for heaven's sake.

Have you forgotten the basic rasterizer architecture elements? Anisotropic is a function of the sampling units- it does not require any additional fillrate. GPUs will take a fill hit IF they lack the proper sampling units to perform the task in a single cycle- if they have the sampling units required for the number of samples needed then there is no fill hit. Also- it consumes bandwidth when there are cache misses- and only then. There was a time when trilinear filtering was a 50% performance hit over bilinear. When the GeForce hit and it had no hit at all everyone assumed it wasn't doing trilinear(trying to bring up an issue you may recall) when in fact it was doing it absolutely correctly- the part simply had the sampling units to handle 8 texel reads per pipe.

SSAA obviously is brute force- no way around that.

The numbers I linked to directly refute there being a cummulative hit. If you were to shave off 60% of 100FPS(8x AF) you would be down to 40FPS- if you then chopped another 60% off for 4xS you would be down to 16FPS- a reduction of 84%. A good deal of the time in GPU limited situations there wasn't a big difference between 4x MSAA and 4xS.

And that means what exactly?

You are far more devoted to them then I will ever be. Of course, I know that you are far more interested in ATi then nV ATM but for some time prior to you buying your GF2MX I have not purchased any nVidia graphics cards(although I have picked up several of their mobos).

Even the R100 was employing non-square sampling patterns in the direction of the stretch.

Some of the time.

Whether you need a non-square sampling pattern depends entirely on each surface in the scene. A good AF algorithm shouldn't apply AF to surfaces that don't need it.

And do you know that is what the NV2x hardware did? It calculated out AF for each pixel on the screen in terms of its footprint and figured out how many samples it needed(then used a weighted blend to combine them). Current hardware does not do this- it completely ignores everything at a given angle- if it needs AF or not.

 

[kylebisme]

Originally posted by: BenSkywalker

First off, anisotropioc means non-square and the highest quality filtering presents a rounded appearance in certian anisostripic filtering tests (which I assume is what you are referring to as "color wheels"); so the other lesser quality methods are not really anisotropy because they look pointy. Is that how your argument flows?

Look up how LOD bias selection works on various hardware- talk to the driver teams about how it is that a LOD bias selection is chosen and ask them if this is fixed or flexible depending on the samples taken and then apply that information to the color wheels. You'll figure it out pretty quick.

Just for laughs I will take you up on that one and come back with the results.

Originally posted by: BenSkywalkerIf there was some sort of major flaw with my simplification of using that BFG would certainly have pointed it out by now- he likes to point out any potential flaw in my arguments. The issue we are debating is how much of an impact this has- BFG considers it minimal while I consider it horrible.

Thanks for the insight on how you vindicate your postions to yourself.

 

[BFG10K]

You are talking about the graphics engine and ignoring gameplay.

LOL, just as I expected.

First it was best possible reproduction.
Then it was best possible reproduction and mathematically provable.
Then it was best possible reproduction, mathematically provable and available.
Then it was best possible reproduction, mathematically provable, available and in realtime.
Now it's best possible reproduction, mathematically provable, available, in realtime with good gameplay.

I'm sorry, please point out where gameplay is mentioned in your original comment "best possible reproduction"?

How can you even continue with this ridiculous charade?

Gameplay is why I buy games- I buy hardware for rendering. I have no clue how your are equating gameplay with graphics

:roll:

So your original demand for "best possible reproduction" doesn't apply to games then? Or does it only apply to games with poor gameplay?

Using this new twisted logic of yours you've excused Sacrifice for not having HDR because it has good gameplay and hence doesn't need to follow your "best possible reproduction" rule above. I would then guess you aren't concerned about filtering quality in that title either? For that matter why even blast ATi for the W-Buffer issue when IQ clearly doesn't matter in Sacrifice as it has good gameplay?

Or are you telling us Sacrifice doesn't have good gameplay, which is why you are so concerned with its IQ?

I can't wait for your new rule addendum for this one. I'm guessing it'll be "best possible reproduction, mathematically provable, available, in realtime with good gameplay except in cases where I BenSkywalker deem to overrule all previous rules whenever it suits me".

Show me a part that has the power to make it playable.

Excuse me? You're the one that has been advocating "best possible reproduction" and now you're turning it around like I made the original comment! That's rich, really rich.

You made the claim and I refuted it by pointing out it wasn't playable. You then produced some flawed benchmarks which I again refuted by pointing out they really couldn't be used to prove anything in the context of your original comment. Now we've come full circle and you're making it sound like I made the initial claim when it was actually you who made it.

Where have I suggested that I want rendering at sub playable speeds?

By demanding "best possible reproduction" and when it was pointed out it wasn't viable because it was unplayable you ended up accusing me of demanding unplayable speeds when in fact you made the inference. I've been saying right from the start it's unplayable and in fact this is the main basis for refuting your demand for "best possible reproduction"

What I have been saying that SLId GTXs have the power to handle it.

They DON'T have the power to handle best possible reproduction which is 2048x1536 with 16xAA. You are now backpedaling because it's been pointed out how flawed your reasoning is.

All over the place in HL2 and lots of them used in FarCry too.

Alpha textures in general are far and few between. You're blowing this way out of proportion just like the optimized AF, not to mention that you're so concerned with alpha aliasing but then turn around and start peddling low/middling resolutions in a feeble attempt to advance your cause.

It is real AA as is MSAA- it is not real FSAA. Mip mapping is real AA.

What the heck are talking about? If mip-mapping is real AA then what purpose does the concept of MSAA/SSAA even serve? Why even invent it?

The only parts offering it have very poor texture filtering at best.

Show me where "best possible reproduction" states texture filtering only. Or is this another "addendum" on your part?

Let me see, that would now make it "best possible reproduction, mathematically provable, available, in realtime, with good gameplay and only AF"?

I don't see how you couldn't possibly understand why that negates any miniscule advantage that anything north of 4x AA @20x15 would provide.

Miniscule advantage? Does that mean you'd have to be d@mn blind to not see aliasing at 2048x1536 with 4xSAA?

And is that miniscule according to your eyes or according to mathematics? And is the gameplay good? :roll:

I never once said anything like it is OK if the game is completely unplayable in fact that is why I am suggesting the $1K plus setup as the option where solid rendering would be viable on today's games.

Nonsense. Your original comment "I want best possible reproduction" has absolutely nothing about performance nor does it have anything about gameplay. These are both two backpedals that you've been forced to perform after it has been pointed out how ludicrous your initial stance was.

You have also shown us 4xSSAA low resolution benchmarks in an effort to prove who knows what. How that scenario factors into "best possible reproduction" I'll never know and I suspect you won't either.

Not remotely close to the best, I was rendering using higher sampling then that with stochastic sampling patterns ten years ago.

Hang on, I thought you stated pre-rendering doesn't count? So which is it Ben? Are we using pre-rendering or not? Or do we only use it when it suits you?

You find numbers with 4xS in use at high res- I can't.

That's your problem, not mine. You asked for best possible reproduction yet you can't find anything of substance to back it up.

The point of posting those was your insinuation that the performance hit for SSAA and AF was cumulative which it isn't.

But it is. Not only that but I never really made the insinuation until after you posted the benchmarks.

Obviously I was clearly indicating when you WERE CPU bound you had overhead to spare.

You won't be CPU bound in my situations, much less in the "best possible reproduction" situations of yours.

Running 1600x1200 in D3 without AA(and with hack job AF) the 7800GTX is still largely CPU limited and pushing just under 90FPS.

1600x1200? There you go again. How is 1600x1200 even a factor when by your definition it's a LOW resolution and therefore not best possible reproduction?

Since 2048x1536 testing was so incredibly commonplace prior to the launch of the GTX.

So you don't have any 2048x1536 numbers, yet you feel you can somehow inductively prove what exactly from low resolution numbers? OTOH I've demonstrated slideshows on even 7800 SLI setups in such situations that you claim are viable. In response to this you turn around and start producing middling/low resolution scenarios like the quote above even while continuing to blast optimized AF for not meeting your demand for "best possible reproduction".

Anisotropic is a function of the sampling units- it does not require any additional fillrate.

A function of texture sampling units and hence it consumes texel fillrate (or rather did on old cards, now it takes it from shader units). One of the reasons the NV25 took such a large performance hit with AF was because it reverted to a x1 part when AF was used. It also consumes memory bandwidth due to texel sampling requirements and that is a fact. So again you're making a mountain out of a molehill; the two methods don't stress the card that differently and for all intents and purposes are cumulative.

The numbers I linked to directly refute there being a cummulative hit

Excuse me? Your own numbers show a cumulative performance hit.

 

[BenSkywalker]

Just for laughs I will take you up on that one and come back with the results.

Feel free. You will find out I am not mistaken.

LOL, just as I expected.

First it was best possible reproduction.
Then it was best possible reproduction and mathematically provable.
Then it was best possible reproduction, mathematically provable and available.
Then it was best possible reproduction, mathematically provable, available and in realtime.
Now it's best possible reproduction, mathematically provable, available, in realtime with good gameplay.

I'm sorry, please point out where gameplay is mentioned in your original comment "best possible reproduction"?

How can you even continue with this ridiculous charade?

When has gameplay ever been a function of a graphics card? What in the he!l are you even talking about? I want the best possible reproduction- GRAPHICS cards are supposed to reproduce GRAPHICS. Silly me, in a conversation about GRAPHICS rendering I stick to those elements. You can say that AoEIII has the most advanvced rendering engine out of any currently available RTS- doesn't change the fact that the gameplay isn't comparable to Sac(one is a historical based RTS while the other is a fantasy based one- extremely different gameplay mechanics).

As far as my argument changing it hasn't in the slightest way- you keep trying to be obtuse and having me explain out elements I would assume would be understood.

So your original demand for "best possible reproduction" doesn't apply to games then? Or does it only apply to games with poor gameplay?

If the game has bad gameplay I'm not too likely to care too much if it is rendered ideally as I won't be playing it.

Using this new twisted logic of yours you've excused Sacrifice for not having HDR because it has good gameplay and hence doesn't need to follow your "best possible reproduction" rule above.

Sac was released in the original GeForce timeframe- how exactly were they supposed to code in HDR support when there was nothing to code it to? It is a DX7 game and despite the fact that graphics don't make up for gameplay- at the time it was considered, along with Giants, to be the most visually impressive games available. It did use Dot3, H T&L and a geometric LOD system.

I would then guess you aren't concerned about filtering quality in that title either?

Of course I am.

For that matter why even blast ATi for the W-Buffer issue when IQ clearly doesn't matter in Sacrifice as it has good gameplay?

It is hard to play Sac when objects keep flickering in and out. With every driver revision I install I try to play it again, it doesn't get fixed so I can't play it. Been sticking with Rise of Nations lately for my RTS fix. Now- as far as IQ goes- the gameplay is enough to get me to buy the game and be extremely pleased with it. That is why I bought the game- to be pleased. The rendering the hardware is doing with the game is not pleasing me. I bought the hardware to render the game. It really is quite simple.

I can't wait for your new rule addendum for this one. I'm guessing it'll be "best possible reproduction, mathematically provable, available, in realtime with good gameplay except in cases where I BenSkywalker deem to overrule all previous rules whenever it suits me".

That is your little made up argument- you can keep that one going if you want- but you are ignoring logic entirely here.

Excuse me? You're the one that has been advocating "best possible reproduction"

It is trying to reproduce real time graphics- if it isn't real time- it isn't reproducing them now is it?

You made the claim and I refuted it by pointing out it wasn't playable.

On the Ti4200 with modern titles which I didn't argue- what I stated is that it would be playable on something with the power of a GTX.

I've been saying right from the start it's unplayable and in fact this is the main basis for refuting your demand for "best possible reproduction"

Right from the start you have been using claims that I never did in terms of making combinations up that wouldn't be playable. I started off my stance on saying real AF is completely viable today(which it is) and that expanded to SSAA being completely viable today(which it is) and once you realized that I was right you started coming up with the most obscene settings you can come up with.

What the heck are talking about? If mip-mapping is real AA then what purpose does the concept of MSAA/SSAA even serve? Why even invent it?

AA only indicates anti aliasing. Mip mapping in terms of screen real estate and degree eliminated a he!l of a lot more aliasing then anything we have seen since. MSAA only deals with geometric edges as you know- SuperSampling is the only full scene anti aliasing solution we have around currently. The degree in which aliasing exists is significantly reduced from the times of 400x300 titles with point filtering- so much so that what we are using it for now is a relatively speaking miniscule portion of aliasing artifacts- how much of a degree they matter to you is all relative. I can't stand texture aliasing- it annoys me enormously- so I place a much higher degree of importance on proper AF then I do any of the AA implementations.

And is that miniscule according to your eyes or according to mathematics?

Actually yes, it is proveable. Due to the dot pitch on 22" displays the end results of sampling an image running @2048x1536 with 4x AA is significantly higher then what the display can possibly show you. Due to this, the outputted image is being sampled at a rate superior to that of any display I have seen and as such at that point your limitation is your monitor- not your graphics card. This is a discussion about grapihcs cards after all.

Nonsense. Your original comment "I want best possible reproduction" has absolutely nothing about performance nor does it have anything about gameplay. These are both two backpedals that you've been forced to perform after it has been pointed out how ludicrous your initial stance was.

GAMEPLAY HAS NOTHING TO DO WITH IT Good gawd man, are you listening? Gameplay has to do with what type of game I am going to be playing- it doesn't impact rendering outside of how much it will be rendered in front of people. In terms of there being any indication about performance- why on Earth would I continue to point out that SLId GTXs was capable of handling it decently today when the Ti4200 came out years ago and could do the same? I have backpedaled from nothing- you are simply creating an argument up as you go along.

Hang on, I thought you stated pre-rendering doesn't count?

Of course it doesn't- but pretty soon you will be talking about RenderMan being significantly superior to anything out now(which is true) and saying that is what I am claiming everything should be. That is the way you are going.

That's your problem, not mine. You asked for best possible reproduction yet you can't find anything of substance to back it up.

Quite clearly it is entirely your problem. You assume that the GTX can handle 4x AA @2048x1536 in the realm of triple digit framerates but 4xS isn't playable- overall there wasn't that large of a performance hit between 4x MSAA and 4xS the overwhelming majority of the time. The numbers I linked to show that fairly decently too.

1600x1200? There you go again. How is 1600x1200 even a factor when by your definition it's a LOW resolution and therefore not best possible reproduction?

Do I have to bring up FFTSW yet again? How many times do I have to tell you that lower resolutions rendered in a significantly superior fashion will look much better then a superior resolution rendered in a considerably inferior fashion? I have mentioned that explicitly numerous times in this thread.

A function of texture sampling units and hence it consumes texel fillrate (or rather did on old cards, now it takes it from shader units).

Texture sampling units != texture management units. There is a big difference. The GF DDR could output 480Mpixels a second point filtered- it could output 480Mpixels with trilinear filtering. It had eight sampling units per TMU. No magic law says they need stop there.

One of the reasons the NV25 took such a large performance hit with AF was because it reverted to a x1 part when AF was used. It also consumes memory bandwidth due to texel sampling requirements and that is a fact.

It only stresses bandwidth if there is a cache miss- if the sample is being stored on die then it isn't touching mem bandwidth.

So again you're making a mountain out of a molehill; the two methods don't stress the card that differently and for all intents and purposes are cumulative.

It depends on the exact situation but there is nothing that will make them stress the same parts of the card in the manner in which you are speaking. It is easily possible to design a part that takes a 50% hit using AA versus straight rendering, and takes a 50% hit using AF versus straight rendering but only incurs a 50% total performance hit when using both.

Excuse me? Your own numbers show a cumulative performance hit.

They show compounded performance hit, not cumulative. That is a big difference.

 

[kylebisme]

Originally posted by: TheSnowman

Originally posted by: BenSkywalker

First off, anisotropioc means non-square and the highest quality filtering presents a rounded appearance in certian anisostripic filtering tests (which I assume is what you are referring to as "color wheels"); so the other lesser quality methods are not really anisotropy because they look pointy. Is that how your argument flows?

Look up how LOD bias selection works on various hardware- talk to the driver teams about how it is that a LOD bias selection is chosen and ask them if this is fixed or flexible depending on the samples taken and then apply that information to the color wheels. You'll figure it out pretty quick.

Just for laughs I will take you up on that one and come back with the results.

Well didn't manage to get anyone from driver teams, but I did get some opinions from some of the more knowledgeable people at B3D. Of particular note is a post from Pete which links to this SGI page which quite clearly demonstrates anisotropy's non-square nature. Regardless, the information is there and I'll just leave it at that.

 

[BenSkywalker]

Well didn't manage to get anyone from driver teams, but I did get some opinions from some of the more knowledgeable people at B3D.

Ask OpenGL Guy(when I picked up my first R3x0 based part I was certain they selecting LOD bias wrong- turns out it was lower caliber base filtering then what I was used to introducing the additional aliasing- anyway OGL explained it to me). The 'ears' as they took to describe them(as good a term as I have heard used) is where the adaptive AF fails. Now they seemed to take issue with the fact that it need not apply AF over the entire screen- but they also pointed out that when you select 16x you will regularly get anywhere from 1x-16x depending on the angle. If you are only getting 1x you are getting isotropic filtering- not anisotropic and certainly not 16x AF.

 

[BFG10K]

When has gameplay ever been a function of a graphics card?

You tell me. "Best possible reproduction" didn't have anything about games or gameplay, it was just a blanket comment that stated you demanded it. Don't try to make it sound I'm bringing up a strawman when in reality it was your comment that started this whole thing.

I want the best possible reproduction- GRAPHICS cards are supposed to reproduce GRAPHICS

But not for games? If not then where exactly is this shimmering problem then? Professional 3D apps? If so you'd be the only one discussing those here.

Of course I am.

But why? I thought games weren't a function of the graphics and therefore your demand for best possible rendering doesn't count?

So which is Ben? Do you want the best possible reproduction in games or not?

I started off my stance on saying real AF is completely viable today(which it is) and that expanded to SSAA being completely viable today(which it is) and once you realized that I was right you started coming up with the most obscene settings you can come up with.

I started coming up with those settings? I? You must be joking. You demanded best possible reproduction and even in this thread you have bashed 1920x1440 as a low resolution. Then when I use your own standards to refute your claims you accuse me of putting those standards up to begin with. Unbelievable!

Actually yes, it is proveable. Due to the dot pitch on 22" displays the end results of sampling an image running @2048x1536 with 4x AA is significantly higher then what the display can possibly show you. Due to this, the outputted image is being sampled at a rate superior to that of any display I have seen and as such at that point your limitation is your monitor- not your graphics card. This is a discussion about grapihcs cards after all.

Wow, so because monitors only support 24 bit colour we don't need the likes of 128 bit colour then? And we don't need a framerate higher than the monitor's refresh rate either?

You know darned well there's much more to this than what a monitor can display and it's quite disingenuous to present a scenario like that to begin with.

Of course it doesn't- but pretty soon you will be talking about RenderMan being significantly superior to anything out now(which is true) and saying that is what I am claiming everything should be.

Nope. You decided pre-rendering didn't count so I kept it off limits. However you keep bringing it back whenever it suits you.

You assume that the GTX can handle 4x AA @2048x1536 in the realm of triple digit framerates but 4xS isn't playable-overall there wasn't that large of a performance hit between 4x MSAA and 4xS the overwhelming majority of the time. The numbers I linked to show that fairly decently too.

(1) Logically there's no way MSAA and SSAA can have the same performance hit. Anand seems to agree. Additionally I don't think 4xS is 4xSSAA much like 8xS isn't 8xSSAA. I'd wager 4xS is actually 2xSSAA + 2xMSAA.

(2) Back to 4xSSAA examples Ben? Clearly you continue to ignore that 16xAA is the best AA around and again you violate your own request for best reproduction available.

They show compounded performance hit, not cumulative. That is a big difference.

No, no different, only semantics; AF + AA is slower than either AF or AA by itself. That's cumulative.

 

[BenSkywalker]

You tell me. "Best possible reproduction" didn't have anything about games or gameplay, it was just a blanket comment that stated you demanded it. Don't try to make it sound I'm bringing up a strawman when in reality it was your comment that started this whole thing.

Reproduction is reproducing that which is already there. If I move to a set of Senn 650s over my current 580s I certainly don't expect it to make older recordings sound better then the newest ones. Purchasing a DVD-A of an album recorded in the 1960s isn't going to do me any good as it can't improve on the very limited source it is dealing with. If I were to purchase the latest Britney Spears album it would benefit from the newer equiptment and technology but it wouldn't change the fact that I would rather listen to cats drowning over a phonograph

But not for games?

Of course for games. The games that I want to play.

So which is Ben? Do you want the best possible reproduction in games or not?

I want the best reproduction in games I play.

You demanded best possible reproduction and even in this thread you have bashed 1920x1440 as a low resolution. Then when I use your own standards to refute your claims you accuse me of putting those standards up to begin with. Unbelievable!

No- the settings you are talking about are not viable for reproducing gameplay.

Wow, so because monitors only support 24 bit colour we don't need the likes of 128 bit colour then? And we don't need a framerate higher than the monitor's refresh rate either?

CRT monitors handle 128bit color just fine. For framerate we can disable VSync.

You know darned well there's much more to this than what a monitor can display and it's quite disingenuous to present a scenario like that to begin with.

Best I can say to you on this is to check out a 2048x1536 display running 4x AA and you will see what I'm talking about. You are significantly beyond the sampling of the monitor already at that point. Moving to 16x AA from 4x AA running 2048x1536 isn't going to make a big difference on consumer displays- if it does at all.

(1) Logically there's no way MSAA and SSAA can have the same performance hit. Anand seems to agree.

Of course, that is why I stated it wasn't "that large of a performance hit". There is a performance hit- but the difference is nowhere near as huge as the one between AF and psuedo AF.

Additionally I don't think 4xS is 4xSSAA much like 8xS isn't 8xSSAA. I'd wager 4xS is actually 2xSSAA + 2xMSAA.

Yes, 4xS is 2x+2x and not 4x SSAA. It is much better then 4xMSAA though.

(2) Back to 4xSSAA examples Ben? Clearly you continue to ignore that 16xAA is the best AA around and again you violate your own request for best reproduction available.

16x AA isn't the best possible reproduction as it isn't possible to reproduce gameplay at that setting with today's hardware.

No, no different, only semantics; AF + AA is slower than either AF or AA by itself. That's cumulative.

Yes it is semantics- but that doesn't change the fact.

200FPS- 50% hit- 100FPS- 50% hit- 50FPS
200FPS- 50% hit- 100FPS- 10% hit- 90FPS

The top one is cumulative- the bottom is compounded. Semantics is rather important on this one.

 

[nemesismk2]

Is this topic back to talking about the Kyro? Guess not!

 

[BFG10K]

Reproduction is reproducing that which is already there.

So why aren't you running out to pick up a pair of 7800s to run 2048x1536 @ 16xAA to follow your own standards since it is "already there"?

Also why aren't you blasting Sacrifice for being mathematically inferior to AOE3 since AOE3 is "already there"?

If I move to a set of Senn 650s over my current 580s I certainly don't expect it to make older recordings sound better then the newest ones.

That's a flawed example and you know it. Using that logic you could say that GPU upgrades do nothing for existing games which is absolute BS and you know it.

No matter which way you spin it you are being inconsistent and aren't practicing what you preach. For this reason you entire beginning premise "I want the best possible reproduction" is flawed. It's nothing more than an excuse to blast optimized AF while trying to look rational at the same time.

Of course for games.

See above. You practice what you preach only when it suits you to bash AF but then dismiss/excuse any other scenario that would also demand you follow the same standards. You don't want best possible reproduction, you just want to trash optimized AF using flawed examples of reduced resolution and/or AA.

No- the settings you are talking about are not viable for reproducing gameplay.

You are talking about those settings, not me. "Best possible reproduction" and "mathematically provable", remember?

CRT monitors handle 128bit color just fine. For framerate we can disable VSync.

Maybe you should of thought along those lines before you made your other allegation.

You are significantly beyond the sampling of the monitor already at that point. Moving to 16x AA from 4x AA running 2048x1536 isn't going to make a big difference on consumer displays- if it does at all.

Using that logic SSAA is useless since the internal image is always four or eight times larger than the sampling limit of the monitor's current resolution. Again you're simply putting a simpleton spin on it just because it doesn't suit your original flawed premise.

Yes, 4xS is 2x+2x and not 4x SSAA. It is much better then 4xMSAA though.

That depends entirely on the sampling patterns.

16x AA isn't the best possible reproduction as it isn't possible to reproduce gameplay at that setting with today's hardware.

And neither is real AF. Are we back to performance again are we? What happened to "mathematically provable" Ben?

Semantics is rather important on this one.

No, they aren't. There might be some separation but it's quite a stretch to claim independant loads.

 

[BenSkywalker]

So why aren't you running out to pick up a pair of 7800s to run 2048x1536 @ 16xAA to follow your own standards since it is "already there"?

Because 7800s have p!ss poor filtering. AA is insignificant in terms of IQ compared to AF.

Also why aren't you blasting Sacrifice for being mathematically inferior to AOE3 since AOE3 is "already there"?

Because from what I've seen is AOE3's gameplay sucks while Sacs rocks. I don't have the full version of AOE3 atm, and from what I've seen I won't be wasting my time either. RoN is vastly superior to AOE3 in terms of gameplay and is along the same lines as it(more of a quasi realistic traditional RTS then the action/fantasy hybrid RTS Sac is).

That's a flawed example and you know it. Using that logic you could say that GPU upgrades do nothing for existing games which is absolute BS and you know it.

No, that is a joke. We don't have the option to remaster existing recordings. We are not taking a set of mathematical equations and reproducing them to optimal standards in order to listen to music- we are just taking what is already there and playing it back. When you have equiptment which can already clearly define the flaws in the source recording then you aren't going to get any better. Actually- the higher end gear you move to at that point the poorer it tends to sound(as you can pick up even more of the limitations of the original source).

No matter which way you spin it you are being inconsistent and aren't practicing what you preach.

How exactly? I'm not buying parts with disgustingly poor texture filtering- how exactly is that not practicing what I preach?

You practice what you preach only when it suits you to bash AF but then dismiss/excuse any other scenario that would also demand you follow the same standards. You don't want best possible reproduction, you just want to trash optimized AF using flawed examples of reduced resolution and/or AA.

Point out the flaw. Give me a playable setting where I am not being equal in my comparisons.

You are talking about those settings, not me. "Best possible reproduction" and "mathematically provable", remember?

Best possible reproduction of gameplay kind of explicitly implies that it need be playable. I know you understand that.

Maybe you should of thought along those lines before you made your other allegation.

Which allegation? The big edge HDR has is contrast- CRTs can handle the advantage of contrast just fine.

Using that logic SSAA is useless since the internal image is always four or eight times larger than the sampling limit of the monitor's current resolution. Again you're simply putting a simpleton spin on it just because it doesn't suit your original flawed premise.

No, the image sampled to the monitor is the same resolution as what is being outputted. The backbuffer is four times the resolution which is also the case with multi sampling. Signal theory or better yet trying it for yourself would demonstrate it quite nicely.

That depends entirely on the sampling patterns.

Like sampling patterns would be how you would fairly compare them obviously.

And neither is real AF. Are we back to performance again are we? What happened to "mathematically provable" Ben?

It is possible it is not enabled. If it were enabled I would be running 7800GTXs in SLI right now.

No, they aren't. There might be some separation but it's quite a stretch to claim independant loads.

So you are on record saying that dropping 40FPS does not matter- if dropping 40FPS doesn't matter then you agree with me that all optimized AF should be disabled. Why have you been arguing this all along if 40FPS is irrelevant?

 

[SonicIce]

for the record, i can't tell the differance between 150Hz with vysnc and 85Hz with vsync, 60fps seems a tinney bit slower but it's very very subtle. i tested in an old game that i know i can get well over 150fps consistantly (UT99).