ATi is ch**t**g in Filtering

Oh, they screwed up. That's alright then. Move along people, nothing to see here...

 

http://www.ati.com/online/chat/

Excellent way to resolve things.

 

Now the dust is beginning to settle, I think ATI have missed a massive marketing opportunity here. They should have came out on day 1 and say 'Hey we have this great new adaptive filtering that's better than anything you get now' followed by some graphs showing increased FPS and equivalent IQ. It would have had more impact than 3dc.

It's a shame that anything they do now will just be viewed as spin.

 

I was going to ask the same thing. I think such optimizations (this one, fast trilinear, even brilinear for quite a lot of cases) are sensible approaches, it's just that filtering control is in the APIs for a reason. Of course introduction into an API takes time, so a way to use them before is fine IMO. But I have seen no indications at all that the IHVs are trying to push those optimizations into an API. There are cases where you invariably want full trilinear, and it would be nice if the application could specify that. In other cases, it may be enough to just make the mip map transitions practically invisible.

The really fishy thing is not those optimizations by themselves, but how they are (or not) advertised.

 

So they admitted its in fact not "full trilinear". Ok.
________
Plymouth Road Runner History

 

Great. How many hours from now ?
________
EASY VAPE REVIEW

 

A few reasons I can think of why this isn't ideal.

First off APIs tend to get frozen! This means that any new whoppee-doo filtering schemes that are developed are not accessible to older games (where older can mean just a few months old).

Second it's a whole load of extra work for developers, and my (albeit limited) experience with the way games handle AA/AF is that support is piecemeal at best. Moreover if each IHV comes up with different whizzo-filters, each game effectively has to be aware of each of these quality setting for each IHV. Nightmare!

Personally I'd much prefer a centralised control panel approach, which can evolve with new drivers. New filtering features can be retro-fitted to older games. Using the new NVIDIA app. specific settings is a nice idea IMO (when it works properly!). People who don't want app-specific settings can just use a global setting.

My €0.02.

 

and where can reach the chat , didn't see any link or is it IRC(what network,channel?)

 

Is the consensus that this "optimization" will be an option in the next driver version?

 

from what I gather, they're going to implement a web-based chat interface in that page, or link to one, anyway.

 

Just seen the email about the chat, I'll make a couple of additions:

 

________
Honda XL175 history

 

I think there is a simple reason for this with X800 - they forgot. Being a 130nm product they lifted some of the code for RV360, and it inherted its texture filtering abilities almost by default. How the driver guys knew but PR / Marketting didn't is another question - however, when I mentioned to PR that 9600 has been doing this from day one the initial reaction was "no, we don't have that capability".

I think it might be simpler than his in the frst place anyway - I think if the developer supplies the mip map levels then it will do exactly as the developer asks; if the developer relies on autogenerated mips then its being left to the hardware/drivers which falls into the realms of "let the IHV decide".

 

ATI are stuck between a rock and a hard place.

If they include an option in the CP then we will see FPS falling all over the place as reviewers try and compare, rightly or wrongly, "apples to apples".

If they stick to their guns and say 'you don't need to switch it off we guarantee equivalent or better IQ' the wrath of the community is apon them!

It's all going to depend on how clearly ATI can show no loss of IQ vs how clearly the community CAN show a visible loss.

At the moment it nudging towards ATI. Time for a Poll!

 

@democoder:

I hope it will not be an option in the next drive release. I hope that you cannot turn it off and the reason for that is the following:

Suppose you can turn it off. Then review sites like Anandtech, Tom's and most others will most likely tell their readers that this feature is something like Nvidia's brilinear and treat it as if it was exactly the same.

They will do performance tests with both ATI's and Nvidia's optimizations both turned off or both turned on. And they will ignore the fact that ATI's method doesn't give visible lower IQ, and Nvidia's does.

I suspect there will only be something like 2 sites that would dare to do high IQ performance tests with ATI's feature turned on and Nvidia's turned off.

I also strongly believe that this is the reason why ATI kept it a secret.

 

If it appears as an option it should say

'Enable High Quality Filtering'

rather than

'Disable Filtering Optimisations'

 

Thats why theres an extension mechanism in OGL in the first place.
So you think that all new texture compression schemes should just "sneak in" in disguise, for instance ? Would it have been OK for ATI to silently apply TruForm whenever the driver got a batch of tris that seemed suitable for it ?
Somehow though various vendor OGL extensions still get developed, and god forbid, some even ratified by ARB as official.
Somehow MS managed to come out with PS 1.4, PS2.0a and b etc etc .. but for this amazing new filtering scheme there was simply no room eh ?

And for "accessible to older games" .. those older games would be pretty happy with vanilla old trilinear or bilinear as it is, after all this is a new benchmark-busting superchip which should have no trouble running them, right ?

This rant is not specific to ATI and this case, im generally annoyed with the way graphics industry is moving. At some point in time we will be in a situation where everything will be rendered as driver developers damn well please, nevermind what the API and application developers specify. Oh, perhaps triple-A titles will even look and perform barely acceptably, but everyone else ... ?

 

One problem of putting something like this in an API is that ATI will probably have to tell what exactly they are doing.

And they might not want to do that, because this feature is giving them a performance edge over Nvidia.

 

Well, I'd like to have benchmarks like this:

Trilinear vs Trilinear
Brilinear vs Brilinear (measures "quality" of Brilinear implementation, and "image analysis" error rate. NVidia's drivers already do a sort of "poor mans" image analysis: it is done by human beings and enabled via application detection vs ATI's "automatic" analysis)
Bilinear vs Bilinear

If there isn't even an option in the driver, it will be harder to analyze the exact IQ differences. It seems they can only compare it to older cards. We need the ability to take screenshots on X800 with optimizations and without, so we can go over them with a fine tooth comb.

I just don't see the point of taking away an option from users just because some reviewers might use it in ways you don't like. The point is to deliver choice to users, not be afraid of benchmarks.

 

a fact? hardly! From what i gather from this thread so far is that nvidias brilinear and ati's brilinear give the exact same IQ. I would like to see some testing with a couple of games where ati or their adaptive alogroithm apply bilinear and where not. I wouldn't be surprised to find out that when you take 20 games and test them to find out that in none of those games the algorithm would apply full trilinear.

Funny thing is that ati is proving nvidia right with this. They were using bilinear over a year now with nv3x and there were a lot of people not so happy about it(me included) and now ati's practically does the same. The difference however when i read some of the people thoughts on different forums about it is that nvidia cheats(lowering IQ) while ati doesn't cheat(not lowering IQ) even though the output is the exact same