Lars looks at Aquamark (German)

[Mark0]

I liked the "We grayed the scores with the NVIDIA drivers v45.23 and v51.75 because they do not render the correct image (see quality comparison)" idea.
May be this is a good "mean" to show the difference between legal/illegal drivers even to the great public?

Bye!

 

[Vortigern_red]

The driver v45.23 does not render the shadows from objects with AquaMark 3, but it renders them in Aquanox 2. We'll talk with Massive to find out if they changed something in the shadows for AquaMark 3.

But not to render objects normally does not result in lower scores. So all results with NVIDIA´s v45.23 and v51.75 are invalid and should be taken as preliminary.

Our speculation is that NVIDIA reduced the filtering quality to achieve a better performance.

We crayed the scores with the NVIDIA drivers v45.23 and v51.75 because they do not render the correct image

The critics have spoken!... GeforceFX 5900 available at major retailers now!

 

[OpenGL guy]

Is the reason the 44.03 drivers offer a sharper image due to Nvidias Texture sharpening? I find with ATI drivers that if no AF is used or trilinear AF is on it's much softer but with Bilinear AF on it's as sharp as 44.03.

From what I've seen, the ATI cards don't use Aniso correctly with AM3 when set to "application" in the control panel, but it works fine if you force it with the control panel. Kind of the opposite of UT2k3. Hence, the blurry textures on some of the ships in earlier reviews. It wasn't due to 45º angles. Aniso wasn't actually working due to having chose "application" in the drivers.

This paragraph is completely wrong. There are some limitations to our AF algorithm, but it is working as designed. The screenshots from the comparision show one bad case and one good case. In the good case we are equal to nvidia's results on the 44.03 driver, on the bad case we are somewhat worse than the 44.03 driver.

I see that Tom's touched on this in the article, only they said it's like this for the Det 51.75's. I wonder if they are aware that it's like this with the 9700's too?

No he didn't, because it's not.

This is the page where he talks about having to force Aniso with the control panel for the Det drivers:

http://www.tomshardware.com/graphic/20030918/aquamark-12.html

Seeing that both cards are having problems with this, maybe it's a problem with the game itself?

No problem with the game/benchmark as far as that goes.

 

[madshi]

There are some limitations to our AF algorithm, but it is working as designed. The screenshots from the comparision show one bad case and one good case. In the good case we are equal to nvidia's results on the 44.03 driver, on the bad case we are somewhat worse than the 44.03 driver.

I've some filtering related questions, would you mind answering them?

• Is this limitation of your AF algorithm a hardware limitation? I guess so, but I'm not 100% sure.
• Do you personally feel well with the limitation of your AF algorithm? Probably you can't speak for ATI, but just your personal opinion?
• Do you see a chance that ATI will get rid of this limitation with the next major core update (probably R500)? Probably you can't answer to this one, but I've asked it nevertheless, just in case...
• Dave found out that the 9600 can do the same quasi trilinear filtering that the GeForce FX does in UT2003 (when you lower the 9600 IQ sliders a bit). But the 9800 doesn't seem to do that. It seems to directly go back to bilinear only. Is that right? Why is that? Actually I do like this quasi trilinear mode - as an additional alternative only, of course.

I'd be thankful for any response - thanks!

 

[Tim]

Rather suspicious:

"The driver v45.23 does not render the shadows from objects with AquaMark 3, but it renders them in Aquanox 2."

 

[jjayb]

This paragraph is completely wrong. There are some limitations to our AF algorithm, but it is working as designed. The screenshots from the comparision show one bad case and one good case. In the good case we are equal to nvidia's results on the 44.03 driver, on the bad case we are somewhat worse than the 44.03 driver.

No, look again at some of the reviews on the web. There is no aniso applied on parts of the scene. At first I thought it was because of the "only apply aniso at certain angles". But I've since realized it doesn't happen when aniso is forced in the control panel.

Here are some screenshots to show you what I mean:

The first is using the control panel set to "application" with 8xAF. Look at the top of the buggy going up the hill on the right:

http://www.nvnews.net/images/news/200309/r9800_4xaa_8xaf_application.png

Now look at this one with 8xAF forced with the control panel:

http://www.nvnews.net/images/news/200309/r9800_4xaa_8xaf_driver.png


See the blurriness on the one using "application"? Notice it's gone with the one that was forced in the control panel?

This was being discussed in this thread at Nvnews:

http://www.nvnews.net/vbulletin/showthread.php?s=&threadid=17994&perpage=25&pagenumber=2

 

[WaltC]

opss... something´s strange here...

@WaltC
I thought about that as well. But the Radeon drivers filter very well in the other situation while the 51.75 does not - and the v44.03 and 45.23 (without lighting) filter correct in the first case. So it does not look like an application issue. I´ll speak to Massive about this... but the guys have a very busy week.

ATI told me that they know about some issues. I don´t think that htis is a serious issue compared to those of the 45.23. The frame where the shot was taken is 3600 and the area I showed is in the lower left position of the shot.

Thanks, Lars. Sounds fair enough...

Regarding the rules... NV made an extra press briefing to announce their new driver optimization guidelines to the press. The slides [H] used were taken from the official presentation. But you´re right, they´re not up on the NV side.

Lars

Yes, that's been my only question about the "guidelines" since they became known--that nVidia seems to acknowledge them barely if at all in the drivers they've produced since then. I guess "guidelines" has to be taken literally here as opposed to them representing a statement of definite intent.

 

[OpenGL guy]

This paragraph is completely wrong. There are some limitations to our AF algorithm, but it is working as designed. The screenshots from the comparision show one bad case and one good case. In the good case we are equal to nvidia's results on the 44.03 driver, on the bad case we are somewhat worse than the 44.03 driver.

No, look again at some of the reviews on the web. There is no aniso applied on parts of the scene. At first I thought it was because of the "only apply aniso at certain angles". But I've since realized it doesn't happen when aniso is forced in the control panel.

Here are some screenshots to show you what I mean:

And here are some to show what I mean:
http://www20.tomshardware.com/graphic/20030918/images/aniso1.png

http://www20.tomshardware.com/graphic/20030918/images/aniso2.png

Looks fine to me.

See the blurriness on the one using "application"? Notice it's gone with the one that was forced in the control panel?

This was being discussed in this thread at Nvnews:

http://www.nvnews.net/vbulletin/showthread.php?s=&threadid=17994&perpage=25&pagenumber=2

I can't explain your results, but I don't see such problems in the pictures posted at Tom's.

 

[jjayb]

The pictures I showed looked the same on all the other reviews I've seen that used that scene. They all showed the Radeon had the blurry top of the buggy. Why does the top of that buggy look blurry?

 

[OpenGL guy]

The pictures I showed looked the same on all the other reviews I've seen that used that scene. They all showed the Radeon had the blurry top of the buggy. Why does the top of that buggy look blurry?

Without looking at the application, I will speculate that the application doesn't specify AF for the buggy. When you use control panel AF, we do AF on all surfaces. When an application enables AF, it's free to enable it on select surfaces.

 

[jjayb]

Thanks for clearing that up for me OpenGl Guy.

 

[THe_KELRaTH]

It's interesting that the discussion is about why 44.03 driver is producing a sharper image than the ATI cat drivers (which looks to me as though 44.03 is bilinear AF v Trilinear AF) yet these were also the same drivers where it was found that Trilinear AF was being forced off when naming the D3D app UT2003. Maybe Nvidia made a typo and accidentally added AM3 to the list too.

Heh.. this made me giggle while browsing a thread on ut2203 app detection:
*** Posted July 24th ***

OMFG! Just when I think things cant get any worse for Nvidia... *rofl*

Boy how wrong can you be