Trilinear Test (Warning 0.5MB PNG)

[cthellis42]

No. First of all, nobody plays still frames. No matter the game, you're going to be moving.

If only Myst were still popular, we wouldn't be having these problems today!

 

[MrGaribaldi]

No. First of all, nobody plays still frames. No matter the game, you're going to be moving.

If only Myst were still popular, we wouldn't be having these problems today!

Not necessarily true...
I seem to recall reading that Minesweeper is still the most played game, and this optimisation does nothing to degrade the IQ, so I don't see the problem...















Sorry, couldn't resist. I blame the red wine for dinner

 

[mikechai]

No. First of all, nobody plays still frames. No matter the game, you're going to be moving.

If only Myst were still popular, we wouldn't be having these problems today!

Not necessarily true...
I seem to recall reading that Minesweeper is still the most played game, and this optimisation does nothing to degrade the IQ, so I don't see the problem...















Sorry, couldn't resist. I blame the red wine for dinner

Hey, I like minesweeper. I play minesweeper almost every day.

note: This is not a joke.
________
DODGE CARAVAN SPECIFICATIONS

 

[MrGaribaldi]

Hey, I like minesweeper. I play minesweeper almost every day.

note: This is not a joke.

So do I, which is partly what prompted the above post...

 

[cthellis42]

I always thought 3dfx's old Adaptive Mine Filtering techniques were a bit shady, though...

 

[lyme]

Just keep in mind that a test application developed by ATI is much less likely to show problems, as the algorithm in question is likely to have been optimized for that test application. Without knowing the actual algorithm used, it's very hard to know the worst-case scenario for this filtering.

No, the application has just been built. Their testing has been done on textures frequently used in gaming environments. The point of this application is to be somewhat of a pathological worst case.

I'm with chalnoth, at least with that result image it is not the pathlogical worst case. They have given results that would be the best of the worst cases. While I can't blame ATI for writing the application they sent to you ver carefully to minimize the differences between trilinear and trylinear, because honestly if I was in their position that is what I would do and I assume nvidia the same.

As for the pathalogical worst case it isn't. The texture they used is one of the better ones to minimize the display of changes in mipmap levels. While a worst case texture would be one that exentuates the change in mip map levels even when full trilinear filtering is being used.

However in the grand scheme of things I don't really care which is which, I just want them to label it as such.

 

[BRiT]

I'm with chalnoth, at least with that result image it is not the pathlogical worst case. They have given results that would be the best of the worst cases. While I can't blame ATI for writing the application they sent to you ver carefully to minimize the differences between trilinear and trylinear, because honestly if I was in their position that is what I would do and I assume nvidia the same.

As for the pathalogical worst case it isn't. The texture they used is one of the better ones to minimize the display of changes in mipmap levels. While a worst case texture would be one that exentuates the change in mip map levels even when full trilinear filtering is being used.

However in the grand scheme of things I don't really care which is which, I just want them to label it as such.

Alright, why don't you post up textures that would be the absolute worst pathalogical case? I'm sick of some saying there are worse cases but not producing what they are. It's time to put up or shut up and stop spreading FUD.

 

[KimB]

Well, for one, since the differences I noticed didn't show up until the far field, a rotated surface would show more differences than that image (due to the angle dependency of ATI's anisotropic degree selection approximation).

 

[Dave Baumann]

We're trying to look at the effects of stright trilinear filtering - we don't want to add other curiosities in place. However, what you speak of is not a texture issue - what texture patterns would highlight the effects better?

 

[BRiT]

But that would merely be due to their implementation of AF and not adaptive trilinear, no?

 

[KimB]

We're trying to look at the effects of stright trilinear filtering - we don't want to add other curiosities in place. However, what you speak of is not a texture issue - what texture patterns would highlight the effects better?

No, it's not a texture issue, but from the screenshot you posted, rotating the plane would make the effects more visible, as the differences, as far as I could tell, were only visible in the far field.

As for what texture patterns would highlight the effects better, I really cannot say, as I have no idea what ATI is doing. If ATI would publish the actual algorithm used, then perhaps it could be proven what situation would be the worst-case for the effect.

But one idea on checking for other artifacts may be to slowly rotate the texture on the surface and see if anything pops up. If ATI released the source code, this should be relatively simple to do.

 

[Dave Baumann]

No, it's not a texture issue, but from the screenshot you posted, rotating the plane would make the effects more visible, as the differences, as far as I could tell, were only visible in the far field.

No, a rotated field would introduce other elements into the equation that we are not specifically looking at here - we are trying to look at differences with Trilinear filtering.

As for it being pushed back, that was a setting issue that has been resolved in the later screenshots taken in the other threads.

As for what texture patterns would highlight the effects better, I really cannot say, as I have no idea what ATI is doing. If ATI would publish the actual algorithm used, then perhaps it could be proven what situation would be the worst-case for the effect.

You don't need to know what ATI is doing - you just need to know what trilinear should do and decide what textures would highlight where anything less than trilinear would present an issue. So, again, waht pattern(s) besides the one here would benefiot most from full trilinear?

 

[ZenOps]

ATi probably won't publish the method they use for trilinear. If its anything like what I've seen so far, it is far closer to a "textbook" trilinear than Nvidia's Brilinear or optimized trilinear.

And ATi won't just realease it just like that it so that NVidia can copy it.

ATi's method does still sound like trilinear... With some sort of "shifted along the horizon" type of scheme. I mean really people, this stuff seems to me more like nitpicking about whether the AA grid is rotated 22.5 degrees or 45 degrees (sure they looks "different" than the textbook AA, but its no less valid an optimization.)

Come to think of it, it would be possible to get better than trilinear quality (not that anyone would notice) by using a backwards exponential calculation. IE: instead of many mapping levels being squished into the horizon, bring more map levels to the bottom of the screen and less at the horizon (People tend to see the mipmap levels on the "ground" and not off into the horizon, where the pixels are so squashed anyways)

 

[KimB]

No, a rotated field would introduce other elements into the equation that we are not specifically looking at here - we are trying to look at differences with Trilinear filtering.

As for it being pushed back, that was a setting issue that has been resolved in the later screenshots taken in the other threads.

But if you're going to make that argument, you might as well not use anisotropic filtering at all. Anisotropic filtering, after all, can help to hide the artifacts from bilinear filtering.

You don't need to know what ATI is doing - you just need to know what trilinear should do and decide what textures would highlight where anything less than trilinear would present an issue. So, again, waht pattern(s) besides the one here would benefiot most from full trilinear?

No, because we still don't know what the optimization is that ATI is using. It could, for example, not just be a trilinear optimization, but something used in conjunction with trilinear and anisotropic filtering.

 

[KimB]

And ATi won't just realease it just like that it so that NVidia can copy it.

There are these funny things called, "patents."

 

[BetrayerX]

Well, I downloaded the image, looked at it....looked almost the same..but couldn't spot what was wrong.

From that I was about to choose the last option.

The I decided to run Photoshop. I enhanced the image, and there was nothing to really notice. So I decided to split the image in 2 and did a horizontal flip on the right side and imposed it over the left side. Using CRTL Z several times you could notice the right side did blend colors more all the way back, so I choosed Right.

Now if you ask me, after that comparison, the left side looks a bit sharper at the end. But to be honest I would have never noticed without an app.

 

[Dave Baumann]

But if you're going to make that argument, you might as well not use anisotropic filtering at all. Anisotropic filtering, after all, can help to hide the artifacts from bilinear filtering.

But that is fundamentally not what we are investigating here - we are looking at Trilinear filtering, introducing other elements will only cloud the issue.

No, because we still don't know what the optimization is that ATI is using. It could, for example, not just be a trilinear optimization, but something used in conjunction with trilinear and anisotropic filtering.

All the screenshots I've shown in the other threads are just looking at Trilinear filtering alone, so we know this is where its occuring.

 

[Althornin]

nm

 

[bloodbob]

No, it's not a texture issue, but from the screenshot you posted, rotating the plane would make the effects more visible, as the differences, as far as I could tell, were only visible in the far field.

No, a rotated field would introduce other elements into the equation that we are not specifically looking at here - we are trying to look at differences with Trilinear filtering.

As for it being pushed back, that was a setting issue that has been resolved in the later screenshots taken in the other threads.

Actually I think it is entirely relevant if we are going to look at IQ.

If we are just looking at the technical differences then it doesn't matter.