When enough is enough (AF quality on g70)

[Sxotty]

As I said previously I saw the shimmering on my x800 TT (curtouesy of Dig e.g. x800pro flashed to x800xtpe) before I moved on to a cheapr card. It really frustrates me that these companies will not allow the user a greater degree of control without downloading 3rd party tools. Honestly it is a sad day for a business when you rely on 3rd parties to enable funcitonality of your product, and just have to hope they give it away so others can use it.

In other words the excuse that "You can download X and get it to work" is not valid to me, I would much prefer I could go into the silly control panel and just do it there with nothing but the drivers and control panel from ATI or Nv installed.

 

[trinibwoy]

Is someone from 3dcenter/ computerbase / hardware.fr looking at the ATi shimmering or is not interesting as it is ATI and not nvidia?

That would be interesting. But probably not.

 

[Ailuros]

That would be interesting. But probably not.

On default there's more shimmering on R4xx's then on R3xx's, because the first contain more filtering related optimizations.

If anyone would go through comparisons with the 78.03 betas, the G70 will present more shimmering with default "quality" but only in some selected applications like UT2k4 (and that not across the entire game, but a few specific textures in specific maps) amongst others.

Since optimizations aren't going anywhere in the nearest future, all NVIDIA needs to do IMHO is to build on the 78.03 changes and further improve "quality" output.

 

[trinibwoy]

Since optimizations aren't going anywhere in the nearest future, all NVIDIA needs to do IMHO is to build on the 78.03 changes and further improve "quality" output.

Given Rys' and Ratchet's results which show very little performance hit with the fix, why is it that Nvidia chose only to apply it to the HQ mode? I'm assuming here that the shimmering in both modes has the same root cause.

 

[Ailuros]

Given Rys' and Ratchet's results which show very little performance hit with the fix, why is it that Nvidia chose only to apply it to the HQ mode? I'm assuming here that the shimmering in both modes has the same root cause.

Hopefully I'll be done in couple of days with the write-up for a few more details (in simple engrish: I don't want to spell it all out yet ).

I suspect that the 78.03 are (hopefully) just a very quick short notice "fix"; shimmering in HQ is not 100% gone either in the most troublesome cases (edit: but it takes some serious screen-nose-glueing to notice).

 

[Sharkfood]

Honestly it is a sad day for a business when you rely on 3rd parties to enable funcitonality of your product, and just have to hope they give it away so others can use it.

Well, at least there is a way for the end-user to get that functionality! I'd much prefer a hidden, cryptic registry value or other way to ensure that when I buy a product, I can have flexibility in *some way* to get what I want out of it.

But, I do agree with you. I find it rather disturbing that drivers are more about website benchmarks than end-user functionality and delivery of a quality experience to the user.

I'm also reminded about just how perverse this whole thing has become. Why are website reviews totally exercising double-standards as far as features, driver settings and hidden settings go?

I specifically remember for MONTHS- GF3 reviews benchmarking and showing screenshots of anisotropic filtering when NO SUCH FEATURE WAS FOUND IN THE NV DRIVER SETTINGS ANYWHERE. This continued on even into the GF4 infancy release. Here you had website reviewers using 3rd party tools/tweakers/regedit to "sell" a product based on this nifty IQ feature that didn't even have a drop-down, slider or mention ANYWHERE in the drivers!

Now.. not only will they not use any 3rd party tools (and pretend, suddenly, that anything hiding in the registry or 3rd party tools "doesn't exist".. flipping a 180), but they will also now decide the IHV dictates "how" the benchmarking should be performed- just how "apples to apples" is NVidia's "Quality" mode compared to (whatever) mode R3xx/4xxx GPU's are being benchmarked at? Of course, this is a pretty big problem!

It occurs to me this entire thread and discussion is less about any particular IHV than it is about website reviewers trying desperately to provide "quantitative" reporting on something that absolutely, positively needs "qualitative" representation. It's a big oxymoron... a paradox.. it doesn't exist!

While we can bicker back and forth about "who" is setting "what" baseline (in this case, NV with their "Quality" designation)... it's less about NV and more about any monkey with a web host that will actually abide or practice this and designate it as being anything that even resembles journalism. If anything, SOMEONE needs to define a baseline of comparison... but this then prompts the other IHV(s) to provide the same in a "competitive" fashion for a more apples-to-apples means of comparison.

How much "value" such comparisons add to the end user is directly proportional to how close to the "average user's needs" such baselines have been created to accomodate. This whole thread is basically about one(1) website screaming bloody murder that, with anisotropic filtering in mind (something designed to reduce texture aliasing, duh), what they're testing doesn't really accomodate this... IN THEIR OPINION.

I don't care if both IHV's create a big, ugly "BENCHMARK MODE" button that greys out all the other tabs of their drivers specifically for websites. Perhaps this is truly what is needed in order to accomodate the spurious and missing complexity and advanced user skills many have illustrated in years past.

 

[KimB]

I specifically remember for MONTHS- GF3 reviews benchmarking and showing screenshots of anisotropic filtering when NO SUCH FEATURE WAS FOUND IN THE NV DRIVER SETTINGS ANYWHERE. This continued on even into the GF4 infancy release. Here you had website reviewers using 3rd party tools/tweakers/regedit to "sell" a product based on this nifty IQ feature that didn't even have a drop-down, slider or mention ANYWHERE in the drivers!

Bear in mind that anisotropic was always enableable through the application.

And anisotropic filtering wasn't designed to reduce texture aliasing, but rather to reduce blurring that occurs through normal bilinear filtering.

MIP mapping and bilinear filtering were designed to reduce texture aliasing.

 

[Rolf N]

Is someone from 3dcenter/ computerbase / hardware.fr looking at the ATi shimmering or is not interesting as it is ATI and not nvidia?

Ehrm ...
The 3DCenter article contains video captures from an X800. Linkage. "AI" on and off are both available.

 

[bigz]

Ehrm ...
The 3DCenter article contains video captures from an X800. Linkage. "AI" on and off are both available.

What version of the Catalyst driver is that though? I've noticed more shimmering in Catalyst 5.8 than I have in Catalyst 5.7.

Edit: According to the linked Hardware.fr article, they're using Catalyst 5.6 and Forceware 77.62.

 

[Sharkfood]

Bear in mind that anisotropic was always enableable through the application.

While API's may have had such options, at the time no games/applications had this (especially none of the games being used for benchmarking or screen grabs). The NV product was the first with this "feature" (despite the feature being completed disowned and unsupported by the IHV enjoying the success of said feature.. it was just a registry "hack" much like some of the similar hacks in the GF1/2 line of lesser degree).

What version of the Catalyst driver is that though? I've noticed more shimmering in Catalyst 5.8 than I have in Catalyst 5.7.

I made comments on Rage3D about this as well. I've noticed a plethora of IQ issues in the 5.8's, so have reverted back to the 5.7's.

Texture aliasing, strange LOD in some games... I've also noticed more z-buffer errors in DX7.0/8.0 games- such as entire trees flickering on/off. I'm unsure if the 5.8's should be trusted for any form of website benchmarking at this point.

 

[KimB]

While API's may have had such options, at the time no games/applications had this (especially none of the games being used for benchmarking or screen grabs). The NV product was the first with this "feature" (despite the feature being completed disowned and unsupported by the IHV enjoying the success of said feature.. it was just a registry "hack" much like some of the similar hacks in the GF1/2 line of lesser degree).

Games/applications always had the capacity to support anisotropic filtering.

What you're complaining about is the lack of a driver option to force anisotropic filtering for games that didn't support the feature (and this forcing of anisotropic filtering and MSAA have lead to their own issues, such as applying anisotropic filtering where it's not needed, dramatically reducing performance).

 

[Hyp-X]

While API's may have had such options, at the time no games/applications had this (especially none of the games being used for benchmarking or screen grabs). The NV product was the first with this "feature" (despite the feature being completed disowned and unsupported by the IHV enjoying the success of said feature.. it was just a registry "hack" much like some of the similar hacks in the GF1/2 line of lesser degree).

G400 had 4x AF support way before Geforce cards...
Also Geforce 1 and 2 supported 2x AF...
Even Radeon came out before GF3 with 16x AF support (altough with quality compromises)

 

[no-X]

I tried to enable AF with G400 (win98, DX8.1, G400Tweak V .006, latest drivers), but I wasn't successfull (no IQ difference). Does anybody have some positive experience with G400 + AF?

edit: wasn't the first gaming card supporting AF S3 Savage 3D?

 

[OpenGL guy]

edit: wasn't the first gaming card supporting AF S3 Savage 3D?

It never had AF, but it did have fast trilinear.

 

[Particleman]

Hopefully I'll be done in couple of days with the write-up for a few more details (in simple engrish: I don't want to spell it all out yet ).

I suspect that the 78.03 are (hopefully) just a very quick short notice "fix"; shimmering in HQ is not 100% gone either in the most troublesome cases (edit: but it takes some serious screen-nose-glueing to notice).

I hope that the 78.03s are just a short fix as well. But I am not as optimisstic as I owned a 6800 last year as well, and the situation was almost identical. nVidia improved the HQ mode and added the negative LOD texture clamp and then just stopped, nothing since. The sad part is I think it is possible to fix the quality mode of most the shimmering without a major hit on performance. There would of course be some performance hit, but the fact that we can't enable any of the 3 toggleable optimizations in HQ is what causes the majority of the hit. But I guess nVidia doesn't want to see ANY fps loss in their all so important benchmark mode regardless of how horrible it looks.

 

[Hyp-X]

Does anybody have some positive experience with G400 + AF?

Well I only tried enabling it in our application and it did work.
I have no idea if it can be forced.
That was a G450 btw, but I don't think it had any new features over the G400.

 

[Ailuros]

My humble 2 cents:

http://users.otenet.gr/~ailuros/7803.pdf

 

[rashly]

Very nice article.

 

[Ailuros]

Thanks rashly

 

[bigz]

nice report