So, Any Questions You'd like to ask ATI about Filtering?

[Bjorn]

I just want some comments on the fact they Ati is telling people to use "mip map coloring" tools for deciding IQ while at the same time have implementations that happens to invalidate the results. (it seems that they are very much aware of the fact that it does)

 

[MrBond]

If so, put em here...

Was this "feature" specifically intended for X800 to have, or was it more of an architectural carryover from the RV series (where it appears to have originated)?

Your own marketing docs tell users to use color mip-map transitions to see the true trilinear in action. However, we now know this is a bit of red herring as color mipmap transitions are about the only time this optimization *isn't* activated. How will you amend your marketing documents in the future to more accurately reflect what is actually being done? (Hint: it might help to clarify why this is not the same approach as nvidia's brilinear)

Since everyone's already asked for a checkbox in the drivers, i won't bother

Edit: Is this a hardware optimization, purely a software/driver optimization , or a mix?

 

[cho]

will ati please if competitor use same image-based analyse filter tech ?

 

[Kombatant]

I just want some comments on the fact they Ati is telling people to use "mip map coloring" tools for deciding IQ while at the same time have implementations that happens to invalidate the results. (it seems that they are very much aware of the fact that it does)

That would be my question as well. I would also like to know why was this optimization in the 9600 series only (and not the 9800 series).

 

[vb]

a) how is this "alternative trilinear" done? (algorithm)
b) how many texture accesses? (best/worst/usual)
c) how can we enable it on color mip maps so that we see for ourselves?

Are we assuming with texture caching or with NO texture caching ( gives vary different results ).

Sorry, got it.

make c) how many texture samples? (best/worst/usual)

I just want some comments on the fact they Ati is telling people to use "mip map coloring" tools for deciding IQ while at the same time have implementations that happens to invalidate the results. (it seems that they are very much aware of the fact that it does)

And, that too

 

[tEd]

I just want some comments on the fact they Ati is telling people to use "mip map coloring" tools for deciding IQ while at the same time have implementations that happens to invalidate the results. (it seems that they are very much aware of the fact that it does)

That would be my question as well. I would also like to know why was this optimization in the 9600 series only (and not the 9800 series).

r9800 can't do brilinear it's a hardware limitation.

 

[nelg]

When was the patent filed?
How often are you pushing up against that 15 % threshold?

 

[Jabbah]

Is this optimisation fully adaptive or on/off.

ie. Is there some threshold of difference between mipmaps at which point trilinear is turned off and "brilinear" turned on, or is the region of blending between mipmaps reduced proportionaly to the similarity between them?

 

[ballero]

I just want some comments on the fact they Ati is telling people to use "mip map coloring" tools for deciding IQ while at the same time have implementations that happens to invalidate the results. (it seems that they are very much aware of the fact that it does)

me too

 

[digitalwanderer]

Hey, I just want 'em to start returning me e-mails again!

Seriously, it's starting to get me worried. I haven't gotten the silence treatment out of ATi in years....literally!

 

[Quitch]

Question: Even though NVIDIA and ATI are both using angle dependent anisotropic filtering now, why are some reviewers noting that the AF quality on the NV 6800U card is slightly sharper/clearer at times than the AF quality on the X800 cards?

Did these reviews all turn off tri-ops? Otherwise, it may be that nVidia have brilinear on, while ATI's algorithm has chosen something between brilinear and trilinear, which in turn would lead to nVidia having the sharper image.

 

[Lezmaka]

Are colored mipmaps being detected, or is that a side effect of the "adaptive" stuff?

c) how can we enable it on color mip maps so that we see for ourselves?

Or is there another explanation that the X800 reverts back to trilinear when someone tries to assess filtering quality with coloured mip-maps?

If you read the statement, they explained it:

It works by determining how different one mipmap level is from the next and then applying the appropriate level of filtering. It only applies this optimization to the typical case specifically, where the mipmaps are generated using box filtering. Atypical situations, where each mipmap could differ significantly from the previous level, receive no optimizations. This includes extreme cases such as colored mipmap levels, which is why tests based on color mipmap levels show different results.

Seems to me it's pretty clear that if there's a large difference between the mip maps, there's no optimizations. And with colored mip maps, there seems to be a rather large difference between them, at least to me, say going from bright green to bright red or whatever.

 

[AAlcHemY]

Question: Will there be any new features of this kind in the near future?

 

[tEd]

Seems to me it's pretty clear that if there's a large difference between the mip maps, there's no optimizations. And with colored mip maps, there seems to be a rather large difference between them, at least to me, say going from bright green to bright red or whatever.

The problem is that as by definition mipmaps are always the same texture in smaller size which means that the optimization always kicks in on real gaming situations and only gets disabled when colored mipmaps are used. If you think about it it could just be a clever way to get around detection for colored mipmaps in the first place.

 

[Xmas]

Did these reviews all turn off tri-ops? Otherwise, it may be that nVidia have brilinear on, while ATI's algorithm has chosen something between brilinear and trilinear, which in turn would lead to nVidia having the sharper image.

There is nothing "between" brilinear and trilinear. You can see trilinear and bilinear filtering as the two border cases of brilinear.

blendfactor = (frac(LOD) - 0.5) * x + 0.5, clamped to [0, 1)
x = 1: trilinear
x -> inf: bilinear
everything in between: brilinear

 

[Dave Baumann]

Note - I've already sent a batch of Q's, so more might not get added now.

 

[Maintank]

Why after watching Nvidia stumble and get destroyed in the press would you not disclose something like this?

 

[Stryyder]

Why after watching Nvidia stumble and get destroyed in the press would you not disclose something like this?

Couple of plausable reasons.

1)Comple misunderstanding of the consumer sensativity to the issue.

2)Intellectual rights issues/patents etc

3)Poor internal communitcation

4)stupidity

 

[digitalwanderer]

5)Fear

6)Hubris

 

[Padman]

I think option 3 (poor communication) is most likely. It's not like marketing departments are in the know about every technical detail.

just my 2 cents