AquaMark 3 evaluation, benchmarks and IQ comparisons

[Hanners]

I've finally finished my own article examining Massive Developments attempt to enter the benchmarking market with AquaMark 3, as well as having the opportunity to put both a Radeon 9800 Pro and GeForceFX 5900 through their paces with it.

To take a look at what I thought of both the benchmarks and the cards in question, head over to the article here (http://www.elitebastards.com/page.php?pageid=2086&head=1&comments=1). As always, all comments (bad or good) are most welcome.

[XForce]

Very nice read, thanks for the heads-up! :)

One small question:
on the FX screenshots, the borders of the plants (that I understand are made using alpha-texturesare) are smoothed, but they aren't on the Ati screensshots.

Do the drivers and/or the benchmark force different modes of AA to the various cards (in addition to the 3x limit for Ati cards you mentioned)?

Thanks & cheers,
Mac

P.S. Good thing I'm not a reviewer, as I'd use the 44.xx dets for speed measurement only and the 45.xx or better 51.xx for IQ comparison only.. :angel:

[The Baron]

What's the LOD bias on each card? The sharpness difference in the textures looks odd to me.

[lordtoran]

Originally posted by XForce
One small question:
on the FX screenshots, the borders of the plants (that I understand are made using alpha-texturesare) are smoothed, but they aren't on the Ati screensshots.

Do the drivers and/or the benchmark force different modes of AA to the various cards (in addition to the 3x limit for Ati cards you mentioned)?


ATI uses pure multisampling FSAA as opposed to the Geforces, and thus can't smooth alpha textures, but there are workarounds used by many games.

[The Baron]

Originally posted by lordtoran
ATI uses pure multisampling FSAA as opposed to the Geforces, and thus can't smooth alpha textures, but there are workarounds used by many games.
Uhh... the GF cards since NV20 do pure multisampling unless you use one of the *xS modes, or so I thought.

[G6-200]

With regard to the AA settings, it makes sense to me that ATI would have three levels and nVidia four, ATI provides 2x/4x/6x AA while nVidia has 2x/4x/6x/8x AA available.
The benchmark is not AFAIK producing the AA, just enabling what is available in the driver.
This would also explain the AA'ed alphas seen on the GFFX, level four AA in the benchmark is using 8xAA, which has a supersampling component.

G6

[The Baron]

Originally posted by G6-200
This would also explain the AA'ed alphas seen on the GFFX, level four AA in the benchmark is using 8xAA, which has a supersampling component.

8x != 8xS. (http://www.nvnews.net/vbulletin/showthread.php?s=&threadid=15821)

[lordtoran]

Originally posted by The Baron
Uhh... the GF cards since NV20 do pure multisampling unless you use one of the *xS modes, or so I thought.

I thought they used a mixture of MSAA and SSAA in some modes, but SSAA by default. The inability of smoothing alpha textures is not the fault of ATI's HW architecture itself however, but relies on the principle of MSAA itself. There are simple workarounds like applying alpha blending instead of alpha checking, but some developers still have not heard of it.

[The Baron]

No, they use MSAA by default--the 4xS mode is 2xMSAA with 2xSSAA, if I remember correctly, and 8xS is 4xMSAA and 2xSSAA.

[Hellbinder]

Originally posted by G6-200
With regard to the AA settings, it makes sense to me that ATI would have three levels and nVidia four, ATI provides 2x/4x/6x AA while nVidia has 2x/4x/6x/8x AA available.
The benchmark is not AFAIK producing the AA, just enabling what is available in the driver.
This would also explain the AA'ed alphas seen on the GFFX, level four AA in the benchmark is using 8xAA, which has a supersampling component.

G6
Now this Guy Got it goin on.. !! :afro:

[lordtoran]

Originally posted by The Baron
No, they use MSAA by default--the 4xS mode is 2xMSAA with 2xSSAA, if I remember correctly, and 8xS is 4xMSAA and 2xSSAA.

I didn't know they had so many modes :D Didn't know of it because my last (and only) nVidia card was a Geforce 2. It had only SSAA and slowed down to half the framerate with 4xAA ...

@Hanners: This is probably the best review pointing out the FX'es weaknesses I ever read. Great job! :p

[Sazar]

I linked you up Hanners...

click on my website link for hardware related news :)

[Hanners]

Originally posted by The Baron
What's the LOD bias on each card? The sharpness difference in the textures looks odd to me.

Didn't touch the LOD bias on either card. I think the sharpness difference is due to the poorer AF on the 44.03 drivers - The 45.23s looked much more comparable to the 9800 Pro in that department, but seeing as those drivers weren't even rendering the whole scene I couldn't possibly justify using them for image quality comparisons.

[The Baron]

Originally posted by Hanners
Didn't touch the LOD bias on either card. I think the sharpness difference is due to the poorer AF on the 44.03 drivers - The 45.23s looked much more comparable to the 9800 Pro in that department, but seeing as those drivers weren't even rendering the whole scene I couldn't possibly justify using them for image quality comparisons.
I figured you didn't touch it... but don't different drivers have different defaults?

Then again, I'm not too familiar with AF changes after 44.03.

[Hanners]

Originally posted by XForce
Do the drivers and/or the benchmark force different modes of AA to the various cards (in addition to the 3x limit for Ati cards you mentioned)?

It's a bit strange really - The benchmarking modes offer the usual range of AA you would expect (up to 6x on the 9800 Pro, I forget how far it goes on the 5900), but when it comes to the automatic screenshot grabbing you have to select the AA mode from the main options screen - All it offers are non-maskable modes up to 3x for the Radeon and 4x for the 5900. I'm hoping to find out more about the whys etc from Massive themselves, but haven't so far.

[StealthHawk]

Originally posted by The Baron
I figured you didn't touch it... but don't different drivers have different defaults?

Then again, I'm not too familiar with AF changes after 44.03.

I think he meant the poorer AF of 51.75, not 44.03.

[Hanners]

Originally posted by StealthHawk
I think he meant the poorer AF of 51.75, not 44.03.

No, I didn't use the 51.75s at all, just 44.03 and 45.23. The AF quality for the 44.03 drivers were noticably worse.

[StealthHawk]

Originally posted by Hanners
No, I didn't use the 51.75s at all, just 44.03 and 45.23. The AF quality for the 44.03 drivers were noticably worse.

Do you have any screenshots of this(I did not see any)?

Because the AF quality between 44.03 and 45.23 seem the same based on what Toms (http://www6.tomshardware.com/graphic/20030918/aquamark-10.html#anisotropic_filtering_quality) showed. The sharpness of the ground textures looks the same, with the apparent difference in lighting being the major change.

[Hanners]

Originally posted by StealthHawk
Do you have any screenshots of this(I did not see any)?

Because the AF quality between 44.03 and 45.23 seem the same based on what Toms (http://www6.tomshardware.com/graphic/20030918/aquamark-10.html#anisotropic_filtering_quality) showed. The sharpness of the ground textures looks the same, with the apparent difference in lighting being the major change.

I probably do have some of the 45.23 screenshots still saved, I'll knock up a little comparison GIF between the 44.03s and 45.23s and post it up here later if I have.