Good explanation of filtering (must read for beginners)

[ChrisRay]

Care to Clarify? From Everything I have read, Changing to an r300 device ID forces the game to run in Full Precision. And running the r300 in FX device ID forces the game to lower Quality. (Tho precision doesnt change) Forcing the game to use Shader 1.1, (However lighting used for normalization of cubemaps and some shaders are switched to 1.1) Rather than 2.0.

And from the Tests I have seen it defaults to FP32 (with the exception of 3dmark) if PP is not requested, I havent seen a DX title yet that doesnt request PP,

If I'm wrong. Provide evidence. Dont tell me I am wrong and to "look" it up. There are some pretty substancial differences between the rendering pathways That are optimised for the NV3x and the r300.

 

[jvd]

Care to Clarify? From Everything I have read, Changing to an r300 device ID forces the game to run in Full Precision. And running the r300 in FX device ID forces the game to lower Quality. (Tho precision doesnt change) Forcing the game to use Shader 1.1, (However lighting used for normalization of cubemaps and some shaders are switched to 1.1) Rather than 2.0.

And from the Tests I have seen it defaults to FP32 (with the exception of 3dmark) if PP is not requested, I havent seen a DX title yet that doesnt request PP,

If I'm wrong. Provide evidence. Dont tell me I am wrong and to "look" it up. There are some pretty substancial differences between the rendering pathways That are optimised for the NV3x and the r300.

I'm telling you to look it up becaue i'm not running a search for something that has been discussed here many times before just because you feel like ignoring things .

The driver forces pp hints in some cases and also the game forces diffrent rendering paths for it (as it wont run at a decent speed)

Search about the demo.

Its late and I'm off to bed chrisray .

 

[ChrisRay]

When you wake up tommorrow you can perhaps find the links for me. It's not like this discussion is going anywhere,

I'm all for learning something. If that is indeed the case. But I dont know what to look for. Since you're correcting me. You obviously do.

 

[Ailuros]

Personally if I'm spending =/>400$ for a graphics card I won't do it for benchmark scores in order to participate in juvenile pissing contests, but rather to get the highest image quality possible.

I couldn't care less which IHV optimizes where, how and to what degree as long as I can enable the highest possible quality on each board. There an added option in the driver control panel is welcome.

I use the latest RadLinker for the current R300 not only because of the application specific presets it allows, but also for it's forced trilinear option; there are of course cases where forcing trilinear on all texturing stages isn't really necessary, yet there are cases where it is at least for me. And that's merely just one example.

I am NOT in favour of either 3rd party applications or registry hacks; I'd rather much more prefer to have as much as possible in the driver. If optimisations increase, then I don't see why driver CP options shouldn't either.

So far I haven't found much evidence yet to applaud ATI when it comes to texture filtering for it's last generation products. Texture filtering quality gradually increased from R100 over R200 and R300. If someone on the other hand can possibly show me any factual improvements in R420 I'm all ears, because I'm obviously blind. In this case optimisations for higher performance with acclaimed similar IQ doesn't count; it's not an IQ improvement after all. And I can't repeat often enough that I don't like the fact that NVIDIA adopted angle-dependancy either.

Finally if I'd attempt to play devil's advocate here, NV adopted angle-dependancy and ATI adopted a similar "brilinear" technique. If I'd shrink all possible optimisations down to only those two, then NVIDIA has actually done the right thing for providing a "trilinear optimisations OFF" switch in it's driver CP and yes I would expect ATI to do the same.

 

[bloodbob]

Okay first of all I want to say I very much like this article this is some very good work I'm generally not overly fussed with the work that comes out of THG but this is a surprise. While I still haven't read the entire article I think the nail has been hit on the head. Both parties are optimising without allowing the user or the developer ( well sensible control it ) to control it on atleast some of their cards.

 

[Diplo]

Except I see no difference in image quality with what ATi did.

What comparisons have you personally done to arrive at this conclusion? Just curious...

Really, though, the point isn't so much about whether IQ has been effected but the fact that ATI deliberately covered up the fact they weren't always using full trilinear when they said they were. What is more damning for me, though, is the fact they actively encouraged reviewers to disable filtering optimisations on Nvidia cards in benchmarks (without ever mentioning their cards were using 'hardwired' optimisations themselves). Even worse they actively encouraged reviewers to only compare IQ when coloured mitmaps were being used, knowing full well this was the only case when their cards did use full trilinear.

Put these two facts together and you can't but help think they were at best being deceptive and at worst plain cheating. The fact that no one realised this until now is a testimony to ATIs skill at optimising filtering, but that doesn't make up for the fact they weren't totally honest with us. If they'd said up front that they were using optimised trilinear and given an option to turn it off (even if it was a registry hack) then I would have zero problem with what they have done since the optimisations appear to be well implemented.

It may seem petty and churlish to admonish them for this 'small' transgression, but this is all part of the slippery-slope/thin-end-of-the-wedge. If we don't put our foot down and demand vendors to be honest with us they will continue to mislead us. The fact that Nvidia may have mislead us more in the past is simply no excuse for letting ATI get away with essentially the same behaviour but better executed.

 

[phenix]

What this means is we make it clear to ATI what we want in a CIVILIZED manner. If everyone floods ATI with emails asking for clarification or options to select/deselect the option (or even a slider), then they would be stupid not to respond, for fear of alienating customers.

Well, its very Marge Simpson of you but thats not gonna happen.
People flooded ATI with civilized mails before on requests for SSAA or full trilinear filtering on all texture stages in control panel. What did ATI's driver team replied? "No we cannot give you that but we will give you something that you will be pleasantly surprised on our next driver release." And what did ATI users get instead (including myself) trylinear.

I am not trolling so dont even think about bashing me. I just wanted to say nothing will ever change for "good and better" in this industry. I will not be surprised if Nvidia steps back from "optimisations off" feature in their next driver and adopt trylinear instead.

my two cents.

 

[Bambers]

ATI are converned about the number of options in the control panel, so they don't seem keen on adding another checkbox. My suggestion to them has been to just keep this as the default option but have it as one notch down the slider, with the full notch as standard trilinear. I think there is some feeling their end that this is tantamount to a "remove-some-performance-for-no-IQ-gain" option, so they weren't too keen on the idea initially. However, if people can spot the differences and they are highlighted hopefully they will recant on this.

thats pretty much what the 8500 used to do in the earlier catalyst drivers. default setting was quality which appeared to use a similar adaptive technique to the one for 9600/x800, moving to high qulaity disabled it.


Another option would be to dump it in the compatibility settings?

 

[Hanners]

Another option would be to dump it in the compatibility settings?

Even a hidden registry setting a la the way texture stage optimisations can currently be disabled would be a start.

Let's be honest, the people who are concerned about this are the real enthusiasts, so they'll have no trouble digging through the registry or making a small tool (or adding it to an existing tweaker) to enable/disable it.

 

[Dave Baumann]

Personally I'd rather not have to dig through registry settings to get full Trilinear in a review - too easy to forget. What we need is some kind of control panel with different user levels "Beginner / Office User" / "Tweaker" / "Extreme knob twiddler Enthusiast" - all of these types of options can be dumped in the higher level control panel interface and for those that really want to dig through them.

 

[Hanners]

Personally I'd rather not have to dig through registry settings to get full Trilinear in a review - too easy to forget.

This is true, although I'm sure an rTool-esque tweaker would soon be created by someone to make a quick and dirty way of enabling it. I'd rather have it integrated into the control panel too, but I just get the feeling it won't happen any time soon, so a 'hidden' registry key is something of a compromise.

What we need is some kind of control panel with different user levels "Beginner / Office User" / "Tweaker" / "Extreme knob twiddler Enthusiast" - all of these types of options can be dumped in the higher level control panel interface and for those that really want to dig through them.

Yep, definitely sounds like a good idea to me.

 

[Quitch]

So have people come to the conclusion that trylinear is worse, quality wise, than trilinear? I must have missed the evidence. If there isn't any, then doing a trilinear to trilinear comparison for the purpose of "apples to apples" is a nonsense, and would be a mockery of a review.

 

[Ailuros]

I haven't had the chance yet to have an extensive look over try- or bri-linear or whatever they call it these days, but I can assure you that there are more than just a few cases where I can see a difference with texturing stage optimisations and to that pretty annoying ones. The majority doesn't notice those either.

IMHO - and that completely irrelevant of IHV - cases where an optimisation or approximation is by 100% on par with what the optimisation is trying to "emulate" are extremely rare. In most cases if not all there will always be at least minor sacrifices and it's really up to the final user to decide what annoys him or what he can tolerate.

As I said before the job of a reviewer is getting harder to the day and it should be close to impossible with whatever approach to escape criticism.

What we need is some kind of control panel with different user levels "Beginner / Office User" / "Tweaker" / "Extreme knob twiddler Enthusiast" - all of these types of options can be dumped in the higher level control panel interface and for those that really want to dig through them.

Definitely. Although even an "Advanced" tab somewhere would be enough already.

 

[Bjorn]

Definitely. Although even an "Advanced" tab somewhere would be enough already.

I'd rather have a "Extreme knob twiddler Enthusiast" tab, it has a nice ring to it

 

[digitalwanderer]

I'd rather have a "Extreme knob twiddler Enthusiast" tab, it has a nice ring to it

OOOOOH! Enthusiastically seconded!!!

 

[Ailuros]

Definitely. Although even an "Advanced" tab somewhere would be enough already.

I'd rather have a "Extreme knob twiddler Enthusiast" tab, it has a nice ring to it

Engineers have a sense of humor?



If you can come up with an equivalent tab name for inexperienced users, I think I could actually like the idea

 

[Stryyder]

My Favorite Quote from that article...

The statements from ATI are, in part, incorrect. True trilinear filtering is mathematically defined and thus almost a standard. Brilinear is something else, i.e. optimized, and thus no longer "true." It is beyond us how ATi can manage to say that this filter is supposed to result in better quality than real trilinear. Moreover, they avoid the question - understandably, from a marketing point of view - of what they think of the benchmark values from various reviews. Several reviewers have disabled the trilinear optimization in NVIDIA's GeForce 6800, as suggested by ATI. In our X800 test NVIDIA's trilinear optimization was not disabled, so the comparable values continue to be valid and comparable.

1) Is it mathematically defined..uh no
2) almost a standard =oxymoron
3) It is beyond us how ATI can manage to say that this filter is supposed to result in meter quality than real trilinear - I guess that shouldn't suprise me but Lars weren't you in a half dozen threads here about how there would be smoother transistions because of this or am I dreaming.
4)In our X800 test NVIDIA's trilinear optimaztion was not disabled (big fucking suprise)

the fact is that ATi's X800 currently cannot handle trilinear filtering and NVIDIA's GeForce 6800 delivers a theoretically better filter quality, at least when the trilinear optimization is disabled in the driver

So much wrong with this statement don't know where to begin....

I must give Lars credit however for a very good explanation of the technology but the conclusions are laughable

 

[Bjorn]

Definitely. Although even an "Advanced" tab somewhere would be enough already.

I'd rather have a "Extreme knob twiddler Enthusiast" tab, it has a nice ring to it

Engineers have a sense of humor?



If you can come up with an equivalent tab name for inexperienced users, I think I could actually like the idea

I didn't come up with it, ask Dave

 

[Operation Mindcrime]

It amazes me how every time this subject comes up, the thread always becomes an IQ issue. ATI cheating is not a matter of IQ but rather one of FPS. The current crop of cards are so close that the only way the manufacturers can distinguish themselves is through minor feature differences and FPS. ATI unquestionably set benchmarking parameters with the intent to skew the FPS results in their favor. It is for this reason AND ONLY THIS REASON that ATI is being accused of cheating.

Is adaptive filtering a bad thing? NO. However, as an intelligent consumer, I hate being lied to and misrepresentation. It indicates to me that the company sees its potential client base as ignorant sheep. It's doubly insulting when a company thinks it can pull this off with a client base full of intelligent IT professionals...

 

[Ailuros]

I didn't like the "Beginner/Office User" that's all

Anyway if I ever find the time I might eventually dip my own foot in the water and attempt to write an article (more of the layman kind) about texture filtering in the foreseeable future.