Hellbinder says full trilinear would cost 50% preformance?

[Pete]

So in other words, We cant take you seriously all the time either. :?

I believe Mufu left his <facetious> tags at home. I can't tell if you forgot your <tongue in cheek> tags in your other pants, though.

Mike, I'm not sure I understand the purpose of those numbers. Could you let me know what you're looking to demonstrate? A "percentage improvement" from one level of AF to another on the same card would seem to be hard to qualify without a test for texture shimmer. I mean, are we seeing more texture detail or more aliasing? Or am I mis/overanalyzing the issue?

I can see the potential benefits of being able to say how much more detail was added, but I'm not sure if it's safe to assume any differences are purely beneficial to the gamer.

 

[MikeC]

Honestly though... Full trilinear is quality overkill for both NVidia and ATi, bilinear isn't really all *that* bad.

I'm discovering a similar situation with these adaptive filtering techniques. The biggest "bang for the buck" is seen when moving from trilinear to 2X aniso. In most situations 8X and 16X aniso are overkill since they don't significantly enhance texture quality over 4X.

 

[Ailuros]

Honestly though... Full trilinear is quality overkill for both NVidia and ATi, bilinear isn't really all *that* bad.

I'm discovering a similar situation with these adaptive filtering techniques. The biggest "bang for the buck" is seen when moving from trilinear to 2X aniso. In most situations 8X and 16X aniso are overkill since they don't significantly enhance texture quality over 4X.

Sad but true. And that's one of the reasons why I'm often mentioning that I would mind much less about angle-dependancy if the amount of samples would get reduced to 4x instead of 2x (it would probably kill the "transistor saving" purpose of angle dependancy though).

Has anyone an idea how feasable it would be to see TMUs in the future (not necessarily next generation at any price) that are capable of at least 2xAF?

 

[ZenOps]

I don't think trilinear was ever free or single pass on any chipset. Even on the newest 6800 and x800, both ATi and Nvidia are trying desperately to find a lower precision than trilinear alternative to save on the extreme performance degradation that true trilinear causes. Add a nice exponential fog ontop of top of a true trilinear, and supposedly the framerates can just kill a videocard.

Nvidia and its Brilinear and trilinear optimized (for 59xx and 6800), now ATi pushing its adaptive trilinear (for 9xxx? and x800)
Its definitely the right way to go, and it is a necessary optimization IMO. There should always be the option to force a full trilinear though, unless there is absolutely no visual difference.

 

[KimB]

Has anyone an idea how feasable it would be to see TMUs in the future (not necessarily next generation at any price) that are capable of at least 2xAF?

The problem with that idea is that even 2-degree anisotropy will go unused a fair portion of the time, since anisotropic filtering chooses the degree to use on a per-pixel basis.

No, I think the answer is simply to have hardware manufacturers work on getting their hardware to better hide the latency of texture filtering by depending on long pixel shader programs, and for game developers to move towards longer pixel shader programs.

After that, all that we need is for hardware manufacturers to not decrease the texture filtering power of the graphics chips, and trilinear/anisotropic filtering performance will take care of itself in time.

 

[Xmas]

Re: Hellbinder says full trilinear would cost 50% preformanc

The oringinal incarnations of AF used the full number of samples everwhere but really you only need a high number of samples at anistropic sampling areas ( upclose its fairly isotropic ).

Which "original incarnations" would that be?

I'm discovering a similar situation with these adaptive filtering techniques. The biggest "bang for the buck" is seen when moving from trilinear to 2X aniso. In most situations 8X and 16X aniso are overkill since they don't significantly enhance texture quality over 4X.

OTOH, the cost for a higher level of AF is decreasing. On average there is only a very small percentage of a scene that has a high degree of anisotropy.

 

[Xmas]

I don't think trilinear was ever free or single pass on any chipset.

Please don't confuse pass with cycle. Trilinear is always single pass, but not necessarily single cycle. Savage2000, DeltaChrome, KyroI/II (only with compressed textures) have "free", single cycle "fast trilinear" which takes all the samples from one mip map. GeForce256 even had free single cycle "pure" trilinear filtering. But there is a bandwidth hit. And on all chips with 2+ TMUs per pipe trilinear is single cycle with single texturing.

 

[Xmas]

However, the higer the level of AF applied, the more you are using samples from just the first mip level in the first place, so the comparative reduction in performance is reduced with AF.

Depends on the AF algorithm used. Some take an equal number of samples from both mip levels, some don't.

Has anyone an idea how feasable it would be to see TMUs in the future (not necessarily next generation at any price) that are capable of at least 2xAF?

The problem with that idea is that even 2-degree anisotropy will go unused a fair portion of the time, since anisotropic filtering chooses the degree to use on a per-pixel basis.

Again, this depends on the AF algorithm used, when AF "kicks in". On NV20 you get at least 2xAF on 99% of all pixels.

 

[ChrisRay]

I believe Mufu left his <facetious> tags at home. I can't tell if you forgot your <tongue in cheek> tags in your other pants, though.

Not sure what you are saying, Just saying if he states that he will intentionally mislead us sometimes. It's hard to take him seriously all the time.

 

[jvd]

I believe Mufu left his <facetious> tags at home. I can't tell if you forgot your <tongue in cheek> tags in your other pants, though.

Not sure what you are saying, Just saying if he states that he will intentionally mislead us sometimes. It's hard to take him seriously all the time.

What if he was misleading u when he said that ?

Besides everyone here misleads you . YOu will do it to us too . Esp to sway us to views .

 

[Tim Murray]

Things are obfuscated at times, yes, because to give every detail (or make detail that he does include as accurate as possible) could easily tip off either company (since just about every board is slightly different to some degree).

 

[Dave B(TotalVR)]

Perhaps the bet game to discuss true trilinear performance issues is counterstrike, as you have the option to force it in the console.

gl_texturemode gl_linear_mipmap_nearest = bilinear
gl_texturemode gl_linear_mipmap_linear = trilinear


I have AF forced for CS, but I have to disable trilinear in the console or else I drop from 85 fps (vsync locked) to 45 fps in certain places on levels which is REALLY annoying as it messes up my whole mouse movement feel if you knwo what I mean.

Im on a 9600 256MB in1280x1024 with 16x AF. Its annoying because I can see the lack of proper filtering on cs_assault (my fave map) when I look at the roof from outside (which is at about 35-40 degrees off the horizontal). Mipmap boundaries are clearly visible and turning on the trilinear solves this, butcosts fps.

Dave

 

[Sxotty]

I don't think we should trash Hellbinder too much. He should just get a sig that says ATI fanboy beware or some such thing . And being a "fan" of a company isn't evil or bad it just tends to bias your arguments and make you try to promote an agenda instead of just enjoying what comes out. I do not know since I haven't seen his posts here lately much but it seems that he might be mellowing (from his former often vitriolic self) even which is good for him if it is so.

 

[MuFu]

Things are obfuscated at times, yes, because to give every detail (or make detail that he does include as accurate as possible) could easily tip off either company (since just about every board is slightly different to some degree).

Eh? No... that was just a piss-take. :?

Pete and PaulS got it.

 

[Tim Murray]

Things are obfuscated at times, yes, because to give every detail (or make detail that he does include as accurate as possible) could easily tip off either company (since just about every board is slightly different to some degree).

Eh? No... that was just a piss-take. :?

Pete and PaulS got it.

I'm covering for you, stupid.

(In case you couldn't tell--it's a joke. A jokey jokey. A Joke of Jokes, if you will.)

 

[ChrisRay]

I believe Mufu left his <facetious> tags at home. I can't tell if you forgot your <tongue in cheek> tags in your other pants, though.

Not sure what you are saying, Just saying if he states that he will intentionally mislead us sometimes. It's hard to take him seriously all the time.

What if he was misleading u when he said that ?

Besides everyone here misleads you . YOu will do it to us too . Esp to sway us to views .

If he was, Just further proves that I cant take his information as reliable. I dont ever go out of my way to "decieve" anyone, So if I do mislead someone. It's never intentional. You can take to the bank.

I think my statement is fair and true, if someone tells you that they will post false information from time to time to mislead people. It comes off as very hard to take what they are saying very seriously.

 

[MuFu]

Chris, that post was a parody. I've never intentionally mislead anybody; I have absolutely no reason to. I rarely even keep information to myself and if I do it's usually out of courtesy towards the source. I'm wrong a lot, but that's not the same thing.

Things are obfuscated at times, yes, because to give every detail (or make detail that he does include as accurate as possible) could easily tip off either company (since just about every board is slightly different to some degree).

Eh? No... that was just a piss-take. :?

Pete and PaulS got it.

I'm covering for you, stupid.

Well do it in a way that actually helps and doesn't put words in my mouth.


Anyways, my apologies, back to the thread topic. Hellbinder, something something 50% performance hit something mathematically impossible banana trilinear filtering something something...

 

[Ailuros]

Has anyone an idea how feasable it would be to see TMUs in the future (not necessarily next generation at any price) that are capable of at least 2xAF?

The problem with that idea is that even 2-degree anisotropy will go unused a fair portion of the time, since anisotropic filtering chooses the degree to use on a per-pixel basis.

No, I think the answer is simply to have hardware manufacturers work on getting their hardware to better hide the latency of texture filtering by depending on long pixel shader programs, and for game developers to move towards longer pixel shader programs.

After that, all that we need is for hardware manufacturers to not decrease the texture filtering power of the graphics chips, and trilinear/anisotropic filtering performance will take care of itself in time.

Requires too much time for my taste in the meantime until we reach that point and you know very well how slow usually developers are in adopting things (and to that not always being their fault).

What I am opposing against here is angle-dependancy to become an industry standard. If I take the NV40 under consideration for the time being it might be able of longer shaders, yet we still have to tolerate a fairly long list of optimisations in order to get fast performing filtering.

I have nothing against focusing on higher degrees of geometry at all, as long as the sacrifices in other departments aren't too big.

 

[digitalwanderer]

I've never intentionally mislead anybody

*COUGH-COUGH-COUGH*
'Scuse me, what did you say?

 

[Blacklash]

How much of a performance hit will ATi take with adaptive AF disabled. Inform us all with your superior knowledge.

I don't think there is any proper way to nail down a number, it will depend totally on the situation and what game is tested.

I'm conducting some off the wall experimentation using AquaMark3, but it's a work-in-progress

http://www.nvnews.net/vbulletin/showthread.php?t=29077

Mike what are your impressions so far? I considered those screens and consider them very much alike(x800/6800u)