R420 and brilinear

[AlphaWolf]

.. but they just "forgot" to give ability of "forcing" AF & TL ... how many people will use 3rd party tools (rtool) ?
so the logic behind some decisions is unclear imho.

So many people whine about this it's funny. Overriding application settings is bad. Complain to developers about lack of support for simple features like AA and AF: IHVs have much more important things to worry about than overriding application settings.

-FUDie

While I agree that developers should be the ones providing access to the features they do not always do so in an unbiased manner. I have seen it mentioned at Rage3d where twimtbp games are apparently disabling features on ATI cards.

Perhaps it would be in their best interest to make it known that these features are available and do work.

 

[YeuEmMaiMai]

I just wish they would offer free trilinear filtering like the Savage3d and 4 did now that would be nice.

 

[FUDie]

I just wish they would offer free trilinear filtering like the Savage3d and 4 did now that would be nice.

It's incredibly expensive, which is why no one is doing it. You're far better off with adding a second texture unit as then you get more global benefit.

Doing single cycle trilinear with a single mipmap requires 16 texture samples... 4 times as many as bilinear. Is it really worthwhile to make your texture unit that much larger?

-FUDie

 

[Xmas]

I may be babeling but I think Hierarchical-Z is part of geometry setup. It is dumping whole polygons before it is reaching rendering engine

Anyway I don't think it works per quad so it doesn't apply.

Hierarchical Z trivially rejects per tile, which may or may not go down to individual quads, depending on the number of hierarchy levels.



vb, I think your definition of "brilinear" does not match with anyone else's definition of "brilinear" here. Because that's what NVidia uses in their current drivers in D3D on GeforceFX cards. Not ATI.

"Brilinear" = a piecewise linear interpolation of mipmap levels, according to the formula
Interpolation factor = (LOD fraction - 0.5) * x + 0.5 , clamped to [0, 1)
where x=1 is plain trilinear and x=inf is plain bilinear, x > 1 is "brilinear".

 

[Dave Baumann]

Hierarchical Z trivially rejects per tile, which may or may not go down to individual quads, depending on the number of hierarchy levels.

Hier-Z is a stage post triangel setup, but pre-rasterisation.

Because that's what NVidia uses in their current drivers in D3D on GeforceFX cards. Not ATI.

ATI has the capability of doing the same thing in their mid range RV3x0 series, but not their high end R3x0 series. If you alter the texture quality slider under D3D with a 9600 you'll see that it is capable of doing the same thing.

 

[Xmas]

ATI has the capability of doing the same thing in their mid range RV3x0 series, but not their high end R3x0 series. If you alter the texture quality slider under D3D with a 9600 you'll see that it is capable of doing the same thing.

I haven't seen any mention of that before. Is it really the same thing, or more like the "one-bit trilinear" ATI introduced in one driver version years ago?

 

[Dave Baumann]

I haven't seen any mention of that before. Is it really the same thing, or more like the "one-bit trilinear" ATI introduced in one driver version years ago?

No, output wise it appears to do pretty much exactly the same. Here's a thread I started about it ages ago.

 

[vb]

Because that's what NVidia uses in their current drivers in D3D on GeforceFX cards. Not ATI.

ATI has the capability of doing the same thing in their mid range RV3x0 series, but not their high end R3x0 series. If you alter the texture quality slider under D3D with a 9600 you'll see that it is capable of doing the same thing.

Thank you.

 

[radar1200gs]

I don't know much about ATI's version of brilinear, so I won't comment on it, however I feel that correctly implimented brilinear filtering will benefit, not hurt the industry.

If nVida and ATI were to replace traditional bilinear filtering with brilinear filtering of the standard seen in the series 5x.xx Detonators while keeping proper trilinear filtering in place then few could argue that the overall quality of 3d graphics hadn't been raised particuarly for older titles that only know about bilinear filtering.

Like all aspects of computing improving the standard of the lowest common denominator can only be a good thing, since it will have the widest impact.

brilinear filtering can also be of value in performance oriented anisotropic filtering, so long as proper trilinear based anisotropic is also available.

 

[StealthHawk]

If nVida and ATI were to replace traditional bilinear filtering with brilinear filtering of the standard seen in the series 5x.xx Detonators while keeping proper trilinear filtering in place then few could argue that the overall quality of 3d graphics hadn't been raised particuarly for older titles that only know about bilinear filtering.

I agree with what you say, but NVIDIA does nothing of the sort. AFAIK their Intellisample settings are still only caps on the maximum IQ, they don't force anything. ie, in games that use bilinear and no trilinear, you will get bilinear instead of brilinear...

I don't see this changing either. Nor do I really see them giving real trilinear back.

 

[radar1200gs]

I never said anything about how nVidia was currently using brilinear, just referenced its quality.

The current lack of control is a major factor why I bought a GF4, not a GF-FX (of course price was another major factor) to replace my GF3.

I'm hopeful that things will improve with the arrival of NV40 though, until proven otherwise I prefer to consider the filtering issues on NV3x to be the result of broken/bugged hardware patched as best as possible, not a deliberate attempt to rip consumers off.

 

[ByteMe]

I'm hopeful that things will improve with the arrival of NV40 though, until proven otherwise I prefer to consider the filtering issues on NV3x to be the result of broken/bugged hardware patched as best as possible, not a deliberate attempt to rip consumers off.

WHAT?????????????

You are not convinced it is all three? (broken,patched,attempt to rip off)

 

[digitalwanderer]

I'm hopeful that things will improve with the arrival of NV40 though, until proven otherwise I prefer to consider the filtering issues on NV3x to be the result of broken/bugged hardware patched as best as possible, not a deliberate attempt to rip consumers off.

I think it's more of a poorly/stupidly/arrogantly designed piece of hardware that they patched as best as possible to rip off the consumers.

 

[Brent]

hardocp has one
http://www.hardocp.com/article.html?art=NTAy

Yes, thanks to Dave which pointed out to Brent in one of many long (I can't remember which) threads about [H] reviewing quality .

Yet, when it came down to bussines http://www.hardocp.com/article.html?art=NTQw suddenly no mention of it, nor anywhere else.

ahem, http://www.hardocp.com/article.html?art=NTQwLDg= under "Driver Tricks & Treats:"

 

[Bambers]

the brilinear in the r200 is not so good as it still produces visible seams. A very brutal version of it is what caused the fuss about non proper trilinear in the first few reviews of the r200.

in d3d it uses graduall more bilinear filtering as the tex pref slider is lowered. In opengl it dropped straight to near bilinear at anything other than max tex pref settings (not default) unless colour mipmaps were on in which case it would somehow flip back to full trilinear. Not sure if it was detecting the contrast between mipmaps or something.