When enough is enough (AF quality on g70)

[FUDie]

Then you're not understanding what I'm saying. Trilinear filtering is nothing more than two bilinear samples averaged together. This is true whether or not anisotropic filtering is enabled. An architecture that could do single-cycle trilinear could also do single-cycle 2-degree anisotropic filtering. But enable trilinear filtering and 2-degree anisostropic, and you'll still get a performance hit from trilinear with this archtecture.

So what you're really asking for isn't single-cycle trilinear filtering, but rather hardware that is capable of more bilinear texture samples per cycle. But that won't solve the issues with trilinear filtering.

Thanks for agreeing with what I said: NVIDIA's only "free" trilinear is bilinear Use brilinear aggressively (as NVIDIA appears to be doing) and you get mostly bilinear.

In any event, I wasn't "asking" for anything. I was merely pointing out why NVIDIA is using brilinear.

Side comment: don't forget that it was ATI that started doing brilinear (albeit less aggressively), and their first implementation of anisotropic filtering didn't even support trilinear.

Newp. Brilinear first appeared on the NV30. And wasn't this original discussion about shimmering on NV40 and G70?

Old games are high-performing anyway, so you can just turn the optimizations off and have no problems.

Now you've gone and painted yourself into a corner. According to you, old games don't need optimizations and new games need more ALU power and therefore don't need texture optimizations. So tell us: Why is NVIDIA going through all this trouble?

Oh wait, all of this is just a bug with the app profiles! Which means that NVIDIA is using heavy brilinear on applications and it can't be turned off. But it's just a bug. But that's what NVIDIA always says.

-FUDie

 

[EasyRaider]

Reviewers should closely analyze image quality and find the closest apples-to-apples settings before running benchmarks. Never mind what the IHV says, reviewer's guides are not to be trusted. Note clearly when a card shows subpar IQ, and if necessary, postpone the bulk of the benchmarks to a follow-up article. That's what I'd do...

 

[KimB]

Thanks for agreeing with what I said: NVIDIA's only "free" trilinear is bilinear Use brilinear aggressively (as NVIDIA appears to be doing) and you get mostly bilinear.

There is no free trilinear, not with anisotropic filtering. Free trilinear doesn't make sense with anisotropic filtering.

In any event, I wasn't "asking" for anything. I was merely pointing out why NVIDIA is using brilinear.

ATI is also using similar optimizations.

Now you've gone and painted yourself into a corner. According to you, old games don't need optimizations and new games need more ALU power and therefore don't need texture optimizations. So tell us: Why is NVIDIA going through all this trouble?

It takes a lot of processing power, and it takes a lot of memory bandwidth to always use highest-quality filtering. These are attempts to improve performance, obviously.

Oh wait, all of this is just a bug with the app profiles! Which means that NVIDIA is using heavy brilinear on applications and it can't be turned off. But it's just a bug. But that's what NVIDIA always says.

Yes, it can indeed be turned off. Just verified myself on my GeForce 6600 GT. Just enable advanced options and you can turn all of the optimizations off just fine.

 

[FUDie]

ATI is also using similar optimizations.

And? ATI is not showing the image quality problems NVIDIA is. Obviously, ATI is taking more care with the optimizations. But this whole discussion was about NVIDIA's filtering issues. Why do you insist on dragging in more irrelevant information?

It takes a lot of processing power, and it takes a lot of memory bandwidth to always use highest-quality filtering. These are attempts to improve performance, obviously.

And obviously what NVIDIA is doing doesn't look good.

Yes, it can indeed be turned off. Just verified myself on my GeForce 6600 GT. Just enable advanced options and you can turn all of the optimizations off just fine.

In all applications? You actually went through every application and verified? Are you using the same driver mentioned on nvnews.net? Also, you didn't test the G70 which appears to show more shimmering than the NV40 even with optimizations disabled so I don't see how your test is relevant.

-FUDie

 

[KimB]

If you have problems with a specific application, you can always modify that application's profile.

And besides, I still am not convinced that brilinear is related to the aliasing issues people are having. It's more likely that it's a result of the new angle-dependent anisotropic algorithm.

 

[FUDie]

If you have problems with a specific application, you can always modify that application's profile.

Like most users are going to know how to do this.

And besides, I still am not convinced that brilinear is related to the aliasing issues people are having. It's more likely that it's a result of the new angle-dependent anisotropic algorithm.

I find that unlikely as there's aliasing on polygons at 0 and 90 degrees where the filtering should be equivalent to the NV40.

As I said before, brilinear can contribute to aliasing because a mipmap's contribution is heavier in areas than it should be.

-FUDie

 

[Blastman]

Like most users are going to know how to do this.

I find that unlikely as there's aliasing on polygons at 0 and 90 degrees where the filtering should be equivalent to the NV40.

As I said before, brilinear can contribute to aliasing because a mipmap's contribution is heavier in areas than it should be.

-FUDie

I don’t understand. Delta Force- BHD (a game I still regularly play), has bad texture aliasing on the desert maps (with no AF). Turing on 8AF (on my 9600XT) gets rid of almost all the shimmering. Whether I run Bilinear or Trylinear (with 8AF) it makes no difference, the IQ is just as good. Bilinear produces just as good IQ -- as far as texture aliasing goes.

 

[KimB]

I find that unlikely as there's aliasing on polygons at 0 and 90 degrees where the filtering should be equivalent to the NV40.

As I said before, brilinear can contribute to aliasing because a mipmap's contribution is heavier in areas than it should be.

Well, here's the issue: brilinear was introduced with the NV3x, and it didn't have the texture aliasing issues that people have been reporting with the NV4x.

I'm willing to bet that the aliasing is due to the positioning of the anisotropic samples, which surely had to change along with the degree selection algorithm.

 

[FUDie]

I don’t understand. Delta Force- BHD (a game I still regularly play), has bad texture aliasing on the desert maps (with no AF). Turing on 8AF (on my 9600XT) gets rid of almost all the shimmering. Whether I run Bilinear or Trylinear (with 8AF) it makes no difference, the IQ is just as good. Bilinear produces just as good IQ -- as far as texture aliasing goes.

Maybe the texture in question had no mipmaps.

-FUDie

 

[WaltC]

Reviewers should closely analyze image quality and find the closest apples-to-apples settings before running benchmarks. Never mind what the IHV says, reviewer's guides are not to be trusted. Note clearly when a card shows subpar IQ, and if necessary, postpone the bulk of the benchmarks to a follow-up article. That's what I'd do...

Exactly--never mind what the IHV says--hits the nail on the head. In fact, if as an IHV you have to instruct users and reviewers to use degraded IQ settings just to benchmark for the sake of frame rate bar charts which you equate to marketing, then I'd say you've got some serious concerns about the competitve performance of your products.

"It's the IQ, stupid," should be the refrain heard 'round the world by IHVs and reviewers everywhere. Every time I think the message has been at long last understood by IHVs, one (in particular) seems to delight in proving me wrong...

Of course, some reviewers are legally blind or else use monitors of such crappy quality that they are effectively legally blind without knowing it--but, hey, if they want to wear the moniker of "reviewer" then they shouldn't shouldn't be afflicted with such problems, imo.

 

[Broken Hope]

Can anyone translate this? Seems like they got an awfully big performance increase from just a few driver releases.

http://www.computerbase.de/artikel/hardware/grafikkarten/2005/bericht_nvidias_g70_texturen/6/

 

[Blastman]

Maybe the texture in question had no mipmaps.

-FUDie

Nope. If I set the CP to application preference (using old CP), the default filtering in the game is bilinear with no AF. Several mipmap transitions on the ground (where the bad texture aliasing occurs) are very visible.

 

[Geo]

Maybe that "we don't have compiler optimizations for specfic apps anymore" statement is slowly going by the wayside as they learn how to take advantage of their beefier ALUs.

 

[tEd]

Can anyone translate this? Seems like they got an awfully big performance increase from just a few driver releases.

http://www.computerbase.de/artikel/hardware/grafikkarten/2005/bericht_nvidias_g70_texturen/6/

All benchmarks done with high quality with a g70!

red bar all optimizations enabled HQ on g70 with 77.77 forceware

blue bar some optimizations enabled HQ on g70 with 76.10 forceware

green bar all optimizations disabled(nv40 quality ?) HQ on g70 with 71.84 forceware

Note: They found a way to install earlier forceware drivers(76.10,71,84) through inf modifications on a g70 even though they don't officially support them.

 

[WaltC]

Maybe that "we don't have compiler optimizations for specific apps anymore" statement is slowly going by the wayside as they learn how to take advantage of their beefier ALUs.

Also, note that "app-specific compiler optimizations" aren't the same thing as "app-specific driver optimizations" in the first place... So, the first could well be true without actually meaning much of anything, imo.

 

[Ailuros]

All benchmarks done with high quality with a g70!

red bar all optimizations enabled HQ on g70 with 77.77 forceware

All optimizations enabled is "Quality" and not "high quality" (albeit I'm not sure if in "quality" upon default the texturing stage optimisation is enabled or not).

blue bar some optimizations enabled HQ on g70 with 76.10 forceware

Still not HQ.

green bar all optimizations disabled(nv40 quality ?) HQ on g70 with 71.84 forceware

That is or should be "high quality".

Note: They found a way to install earlier forceware drivers(76.10,71,84) through inf modifications on a g70 even though they don't officially support them.

I've serious objections when the hardware doesn't get recognized properly. Besides the average 30% between optimisations on and off whatever each bundle includes, isn't exactly a new finding to me either:

http://www.mitrax.de/?cont=artikel&aid=24&page=10

 

[tEd]

All optimizations enabled is "Quality" and not "high quality" (albeit I'm not sure if in "quality" upon default the texturing stage optimisation is enabled or not).

All optimzations enabled regarding g70 on high quality with 77.77 drivers. It's obviously not compareable to nv40 optimisations.

To understand , the article tries to see the performance difference between the addional optimizations added to g70. All benchmarkes are done with drivers to HQ setting but with different drivers which seems to affect the quality output of the AF. Because they way they achieved the results through inf modifications it can't be taken for 100% accurate.

 

[ChrisRay]

As one recognizes on the basis the diagram well, the new ForceWare 77.77-Treiber can set off due to the additional optimizations of the texture filter with an approximate projection/lead from 30 per cent very clearly its competitors with the version numbers 76,10 as well as 71,84. With the two latter the difference is small against it with only somewhat more than two per cent quite. Now one must mark finally however that the material difference between the AF optimizations - thus the distance of the results with all optimizations (77.77), with the "arrow optimization" (76.10) and without any optimizations (71,84, identically to the image quality of the NV4x-Chips) - with all probability smaller precipitates, if the older nVidia drivers recognize the G70 in form of the GeForce 7800 GTX or GT and all chip-internal optimizations (here is not the speech of AF optimizations!) would activate. For this reason those of bench mark are to be regarded rather as appoximate value, which do not become final differences probably so drastically to fail.

Why dont you just wait the Nvidia driver set that fixes it. Rather than drawing incomplete conclusions on an incomplete driver set for the G70 series that clearly wasnt intended for the 7800 cards.

 

[tEd]

Why dont you just wait the Nvidia driver set that fixes it. Rather than drawing incomplete conclusions on an incomplete driver set for the G70 series that clearly wasnt intended for the 7800 cards.

I don't trust nvidia either that's for sure! Why should i? Lied so many times , tried to cheat so many times !

Fuck them!

 

[Ailuros]

All optimzations enabled regarding g70 on high quality with 77.77 drivers. It's obviously not compareable to nv40 optimisations.

There aren't or shouldn't be any optimisations enabled at high quality. All optimisations on for 77.77, means all optimisations enabled.

I do speak German since I grew up in Germany if you've forgotten:

Nun muss man schlussendlich jedoch anmerken, dass der reale Unterschied zwischen den AF-Optimierungen – also der Abstand der Ergebnisse mit allen Optimierungen (77.77), mit der „Pfeiloptimierung“ (76.10) und ohne jegliche Optimierungen (71.84, identisch mit der Bildqualität des NV4x-Chips) – mit aller Wahrscheinlichkeit geringer ausfällt, wenn die älteren nVidia-Treiber den G70 in Form der GeForce 7800 GTX oder GT erkennen und alle chipinternen Optimierungen (hier ist nicht die Rede von AF-Optimierungen!) aktivieren würden. Aus diesem Grund sind die Benchmarks eher als Richtwert anzusehen, die endgültigen Unterschiede werden wohl nicht so drastisch ausfallen.

To understand , the article tries to see the performance difference between the addional optimizations added to g70. All benchmarkes are done with drivers to HQ setting but with different drivers which seems to affect the quality output of the AF. Because they way they achieved the results through inf modifications it can't be taken for 100% accurate.

Then read above, as often as it takes.