Driver Optimisations Continue

lol! priceless! mistakes can be funny!

 

Are there any tools available for image quality comparisons? Is there a way to quantify it I guess is what I'm asking compared to just using your eyes?

 

I'm pretty sure there was a prior discussion about this before. I think ATI's compressionator (I can't spell it) did something along those lines.

Well, presuming that the first version affected was when the profiles were introduced (though, the earlier 60.xx Forcewares series were not tested, only the 65.xx).. according to NVidia's website, 66.93 was released on 11/9/04.. so we're talking about for more than 1 year... approximately 16 months.. so that's a fairly long time.

As for the websites... I'm pretty sure they are just like the normal user.. set it to "High Quality" and "job done". I believe many, if not all websites are affected by this problem.

Good time to look them over once more with the new findings...

 

Oh sure, make poor Hanners work thru his weekend when he's sick.

Meanie.


(Besides, you know he was already planning on ripping into this one this weekend... )

 

Well for B3D reviews it wouldn't be necessary since rarely are they IHV vs IHV, so the optimizations run throughout the line and unless there were comparative differences between generation or model it wouldn't matter since all benefit.

However, those who do compare Ati / Matrox / nV / Xgi should spend the time to ensure that what they are comparing apple to apples. The IQ differences may be minor, but no less than someone declaring a win for 1% performance difference. And considering the effect of this hard to capture difference it's worthwhile mentioning in reviews IMO, even if it doesn't have too much influence on final thoughts.

And DW, really what else is Hanners gonna do; go drinking, carousing and skiing?

 

This sounds like the same bug I made a post about almost a year ago at NV News. It was where all optimisations stayed enabled in HQ mode, even if unchecked, in OGL. The only way to get HQ was to set drivers to quality and disable all optimisations. It must still be broken a year later.

http://www.nvnews.net/vbulletin/showthread.php?t=48096

 

The Compressonator will give you a visual diff (diff = abs(img1 - img2)) between two images. You can bump up the gamma to highlight the differences & also save the diff image. We posted a new version yesterday so your question is well timed.

http://www.ati.com/developer/compressonator.html

GP.

 

I know what you mean; and in fact, lately IQ analysis is almost extinct from major reviews I see. Most of the times I imagine it's because of time constraints - after all, it's quicker/simpler to just let some batch files run and get a few numbers. But I still believe that all IQ options should be set to high because
a) comparing true apples to apples is almost utopian anyway
b) I'd prefer it if I stated "hey, I had all IQ set to max on both vendors, and although SKU #1 performs faster, it does so because the visual quality is diminished. So you, kind reader, make up your mind and know that, when you want the ultimate IQ, you'll never get it with SKU #1, whereas the SKU #2 will give it to you and still be able to run faster if you choose to enable some optimizations"

 

The best way is your eyes, two identical calibrated screens, and two identical PCs except for the graphics card, both running the same game/demo. Then have someone change the settings and randomly swapping the cables.

 

But this does still require any hypothetical site doing the testing to do appropriate checks to actually discover that SKU #1 is diminishing image quality in the first place. If this step is skipped then the conclusion is just going to read "SKU #1 performs better". Under these circumstances (which I believe is likely to be prevalent in reviews following your method) the overall "winner" is most likely to be whichever SKU offers you a high quality mode that doesn't actually allow you to disable all optimizations.

SKU #1 doesn't even need to win all the benchmarks - it might get faster scores overall and hence win a review outright, or it might just gain sufficient extra performance from reducing image quality to cover up the fact that it's really an inferior product, and turn what should be an outright win for SKU #2 into a draw.

It's much easier to test framerates than image quality, and even when IQ testing is done in a review it's a rare occasion where it appears to be anything like adequate for the true complexity of the task. It can also seem a bit of a lost cause - it's not just identifying the areas where SKU #1 looks worse, but its attempting to explain to a largely unaware public _why_ it looks worse, and even why they should be concerned.

 

ATI does it too, bilinear AF for some stages (cat AI on standard), although their filtering is better (less aliasing)

 

I believe that's still true, unless you enable ATI's HQ AF option (only on X1xxx series cards)..

The most uncomparable thing is the game optimizations... where ATI it messed up in an "all-in-one control in CATALYST AI"... NVidia has no method to disable them.

 

I've probably read that before... but read this portion.

Hold on, this was the answer.. but yet this has not be resolved YET??? You're kidding me here... I guess this was not reported to NVidia at the time.

 

I see them (with an X1900) some times with HQ AF too, but not in every case. Would AI on advanced or off completely get rid of it. (cann't try it for now)

 

On what hardware are you seeing this behaviour, or is this entirely second-hand information?

There has never been any "bilinear on some stages, trilinear on others" behaviour on any X*00 or X1*00 part that I am aware of, and f you think this is the case then you are mistaken. There used to be the old "Quality" anisotropic filtering option on earlier control panels for earlier parts (9xxx series), which did trilinear on stage 0 and bilinear on the rest, but this has not been true for any more recent hardware.

 

My own X1900XT, games: SWAT4, FEAR and a techdemo from Inc. (realityengine)
Drivers cat 6.2 beta HQ 16xAF and cat. AI on standard. But I have to say, I'm gfx-whore and have a very keen eye for such things.

 

Hold on, are you enforcing 16xAF from CCC/CP? If you are, that's the problem. You need to use application controlled filtering and set in-game the AF you want.

Why are you using beta CATALYST drivers? 6.2 official has been out a while ago while 6.3 will be out very soon...

From my original statement, I think that's only enforced ONLY when CCC/CP forces AF... not when it is application controlled and probably applies to older hardware (I believe the R2xx series is included)

I believe the trilinear optimization is disabled if you turn CATALYST AI off. You could use a 3rd party tool such as ATI Tray Tools to remove it w/o removing game optimizations.

 

The only way there would be bilinear filtering on any texture stage is if it's set that way by the application, in which case it would also happen with Catalyst AI off.

There will be tweaks to the filtering behaviour enabled with CatalystAI on, and we make every effort to ensure that these tweaks do not produce visible artifacts in the rendered images. If you have a specific reproduction case in these applications then let us know and we will take a look.

Our driver is certainly not forcing bilinear filtering on any texture stage for you.

 

Please report back ASAP...

 

I use the beta because I think there the same as the officials, without a 'stamp'

I'll report back after testing with different AI settings as soon as my rig is finished with some other stuff.