Next B3D 'Tech Article'

[Joe DeFuria]

OK.

I'm about as sick and tired of the "anisotropic filtering debate" as the next guy. Might I make a suggestion to B3D for the topic of the next article: Anisotropic Filtering. (Forgive me if you had some article on this in the distant past....if so, it's time to revive and update it.)

Include:
1) Basic tech background and what it's suppossed to address.
2) Standard terminology for describing quality levels. (nth degree anisotropic). Include description of how the "number of texture samples" fits into it...including trilinear or bilinear filtering.

And then of course, a look at the implementations from several vendors. ATI (8500/9000 and 9700) and nVidia GeForce3/4 of course, but it would be nice to also see Parhelia and P10 if you have access to those boards.

Aside from looking at the "rotating aniso" thing, it would be very useful to look at the "X" quality settings to determine if they are comparable in terms of theoretics. In other words, is nVidia's "2X" setting theoretically comparable to ATi's "2X" setting, and P10's "2X" setting? Do they both represent 2nd degree max anisotropic or something else?

Just thought I'd heap some more work on you guys.

 

[Rodéric]

I want Kristof to finish his shaders article first.

But that would be the perfect one after the Shader Article.

*- edit: spelling -*

 

[Ailuros]

I want Kristof to finish his shaders article first.

But that would be the perfect one after the Shader Article.

*- edit: spelling -*

I second that. Question: did he even start it? hehehe

 

[Randell]

Also a history of when it was first available in consumer boards and the relative importance and performance hit attributable to it. I remember anistropic filtering was claimed as a feature of my S4, but never enabled in drivers as far as I can recall and I've no idea what 'tap' it was meant to be.

AFAIK AF only started to be talked about as a must have feature around the Gf1DDR/GF2 GTS time especiallly by all the 3dfx followers as it was featured in all the Rampage leaked specs.

 

[marco]

We are working on a next article, a true in depth article.
Anisotropic is not in it though. I will talk to dave, wheter he wants to go into it. But we are talking about the next article and may do a co-op writing.
I have 6 pages ready, but I just got some inside info in DirectX.
I won't tell you what's the article about, yet...

 

[T2k]

OK.

I'm about as sick and tired of the "anisotropic filtering debate" as the next guy. Might I make a suggestion to B3D for the topic of the next article: Anisotropic Filtering. (Forgive me if you had some article on this in the distant past....if so, it's time to revive and update it.)

Include:
1) Basic tech background and what it's suppossed to address.
2) Standard terminology for describing quality levels. (nth degree anisotropic). Include description of how the "number of texture samples" fits into it...including trilinear or bilinear filtering.

And then of course, a look at the implementations from several vendors. ATI (8500/9000 and 9700) and nVidia GeForce3/4 of course, but it would be nice to also see Parhelia and P10 if you have access to those boards.

Aside from looking at the "rotating aniso" thing, it would be very useful to look at the "X" quality settings to determine if they are comparable in terms of theoretics. In other words, is nVidia's "2X" setting theoretically comparable to ATi's "2X" setting, and P10's "2X" setting? Do they both represent 2nd degree max anisotropic or something else?

Just thought I'd heap some more work on you guys.

That would be great. 8)

 

[Reverend]

OK.

I'm about as sick and tired of the "anisotropic filtering debate" as the next guy. Might I make a suggestion to B3D for the topic of the next article: Anisotropic Filtering. (Forgive me if you had some article on this in the distant past....if so, it's time to revive and update it.)

Include:
1) Basic tech background and what it's suppossed to address.

Originally, aniso is to make trilinear less blurry for the nearer mipmaps, nothing more, nothing less. A nice side effect is some reduction in aliasing. Shouldn't need an article for this IMO .

2) Standard terminology for describing quality levels. (nth degree anisotropic). Include description of how the "number of texture samples" fits into it...including trilinear or bilinear filtering.

Bilin = 4 point samples. Trilin = 8 point samples. Aniso = 4 or 8 (if bilin or trilin is specified) x aniso degree specified.

Not plugging myself but I think http://www.ve3d.com/3dpulpit/misc/aniso2_gf3/index.html should suffice.

And then of course, a look at the implementations from several vendors. ATI (8500/9000 and 9700) and nVidia GeForce3/4 of course, but it would be nice to also see Parhelia and P10 if you have access to those boards.

This would be nice however. Not sure if it should fall into the "tech article" category. Perhaps better known to be known as "Aniso from various IHVs".

Aside from looking at the "rotating aniso" thing, it would be very useful to look at the "X" quality settings to determine if they are comparable in terms of theoretics. In other words, is nVidia's "2X" setting theoretically comparable to ATi's "2X" setting, and P10's "2X" setting? Do they both represent 2nd degree max anisotropic or something else?

Again, this would be nice. This part can be a "tech article" if explanation is included about aniso pre-filtering in various software/games and how it relates to hardware capabilities.

 

[Joe DeFuria]

OK Rev, here's some examples why I'd like to see an article:

Originally, aniso is to make trilinear less blurry for the nearer mipmaps, nothing more, nothing less.

That's not my understanding. Aniso is to make texture filtering that'sapplied on polygons at oblique angles more accurate. (By taking the samples at more "accurate" directions with appropriate weights.

Bilin = 4 point samples. Trilin = 8 point samples.

Yes, I know this, but it would be nice to have in the article to educate everyone.

...x aniso degree specified.

I'm not sure that's correct. Not sure you just multiply max degree times number of samples in bi/tri linear to get number of samples in anisotropic. (Not that I don't 'trust' you rev, but I would like clarification from another source. ) But if it is correct, at the very least it is at a maximum of "x aniso degree specified. " As, to my understanding, aniso should only increase the number of samples depending on the severity of the obliqueness of the polygon angle to the viewer. (Again, something I would like to see technically explained.)

Not sure if it should fall into the "tech article" category. Perhaps better known to be known as "Aniso from various IHVs".

I just think the two should be combined.

 

[Kristof]

I second that. Question: did he even start it? hehehe

Errr... I think I wrote the title down on a piece of paper, does that count ? 8)

(Seriously its halfway there but needs more... blame the nice british weather for my lack of progress... and yes, I am lazy)

 

[Heathen]

Weather

Well today's weather should give you a good reason to put pen to paper again.

 

[Reverend]

OK Rev, here's some examples why I'd like to see an article:

Originally, aniso is to make trilinear less blurry for the nearer mipmaps, nothing more, nothing less.

That's not my understanding. Aniso is to make texture filtering that'sapplied on polygons at oblique angles more accurate. (By taking the samples at more "accurate" directions with appropriate weights.

More accurate , or simply less blurry? Oh, I see your quotes the second time you mention "accurate" . But, yes, you are "more" correct regarding the oblique angle.

Bilin = 4 point samples. Trilin = 8 point samples.

Yes, I know this, but it would be nice to have in the article to educate everyone.

Yeah, B3D is getting more popular... would be better for this to be mentioned than what I did a looong time ago at 3dpulpit.

...x aniso degree specified.

I'm not sure that's correct. Not sure you just multiply max degree times number of samples in bi/tri linear to get number of samples in anisotropic. (Not that I don't 'trust' you rev, but I would like clarification from another source. ) But if it is correct, at the very least it is at a maximum of "x aniso degree specified. " As, to my understanding, aniso should only increase the number of samples depending on the severity of the obliqueness of the polygon angle to the viewer. (Again, something I would like to see technically explained.)

If you're talking about "programming aniso", it ain't here yet - degree of aniso specified = amount of aniso, period, regardless of "obliqueness". I believe what I said is correct. Of course, getting a second "expert" opinion is no harm.

 

[Xmas]

If you're talking about "programming aniso", it ain't here yet - degree of aniso specified = amount of aniso, period, regardless of "obliqueness". I believe what I said is correct. Of course, getting a second "expert" opinion is no harm.

The aniso level specified in the driver panel is always the maximum level of anisotropy used. All AF implementations adaptively determine how many samples to take depending on how much texels are "stretched" across a surface. It does only indirectly depend on the angle, through perspective transformation.

However, if you want to program AF in the fragment shader, you have to take a fixed amount of samples because of the current lack of data-dependent branching.

 

[Joe DeFuria]

The aniso level specified in the driver panel is always the maximum level of anisotropy used. All AF implementations adaptively determine how many samples to take depending on how much texels are "stretched" across a surface.

That's what I thouhgt.

See Rev, looks like you could use a tech article yourself.

 

[Dave Baumann]

I could knock you all up a nice article on how to lay laminate flooring if you want... Mmmm, might not fit with the sites contents though... Mmmmm, must register BeyondDIY.com...

 

[Randell]

hehe Dave you could have come to me for advice I've laid 2 wooden strip floors and stqined them and 3 laminate floors in recent years The hallway is going to be done professionally, too many awkward cuts etc and a proper wooden floor with nice clear textures None of this laminate approximation stuff

 

[Humus]

Originally, aniso is to make trilinear less blurry for the nearer mipmaps, nothing more, nothing less.

"Nearest" should rather read "fartherest".

 

[SA]

When it comes to anisotropic filtering, I would like to see a 3d hardware implemention using something like Table Feline filtering rather than the more common Texram approach. For a good discussion of these and other anisotropic filtering approaches see:

ftp://gatekeeper.research.compaq.com/pub/DEC/WRL/research-reports/WRL-TR-99.1.pdf

 

[Ailuros]

Errr... I think I wrote the title down on a piece of paper, does that count ? 8)

(Seriously its halfway there but needs more... blame the nice british weather for my lack of progress... and yes, I am lazy)

.....or too busy.

 

[Nick[FM]]

We are working on a next article, a true in depth article.
Anisotropic is not in it though.

Displacement Mapping? At least you said in your Hercules meeting article:

I think either Dave or me will look into this feature later on.

Soooo, it could be it, right?

 

[Gollum]

Thanks for that link SA, truly interesting paper! I'm not sure, but are all current chips using the Texram aproach? The different Feline methods appear to produce a dramatically visible quality improvement compared to that, without burning the extra GPU cycles EWA does. I don't know about the increase in transistor logic, but the paper states it is supposed to be small compared to the improvements in quality. Any idea if some IHVs are working on implementing this?