NVIDIA GF100 & Friends speculation

[CarstenS]

Now, that'd surely need an explanation. As it stands, the results seem incomparable between Radeon and Geforce.

 

[KimB]

It's probably a mistake. One could argue that with the significantly higher quality of transparency multisampling on the GF100, the two options have similar quality. But I think that a more fair comparison would be with neither card enabling transparency AA.

 

[Arnold Beckenbauer]

Does AAA work under DX10/10.1/11?

 

[ShaidarHaran]

It's probably a mistake. One could argue that with the significantly higher quality of transparency multisampling on the GF100, the two options have similar quality. But I think that a more fair comparison would be with neither card enabling transparency AA.

AFAIK it's been written like that in all of their reviews for ages.

 

[Malo]

maybe they just don't understand what the setting means?

 

[Xenus]

It's probably just a case of them not knowing quality means Super-Sampling as opposed to the multi-sampling of Nvidia and simply assuming defaults are comparable.

 

[Kaotik]

According to some googling, at least at some point quality didn't automaticly mean supersampling, but that driver decides wether ss is needed, or should it just use ms.

Could someone test it on some modern title, preferrably one used by xbitlabs bench?
Take just PNG screenshots from some static scene which uses transparency aa, and switch mode and take another, then use compressonator to see wether it really changes things

 

[CarstenS]

Today, under Anti-Aliasing Mode the slider goes from Performance to Quality, but the latter is pure Supersampling, while the former is Multisampling. In between is a setting which is called Adaptive Multi-Sample AA and is the mode where the driver decides.

So basically it could just be a copy and paste bug from older xbitlabs reviews.

 

[Kaotik]

Today, under Anti-Aliasing Mode the slider goes from Performance to Quality, but the latter is pure Supersampling, while the former is Multisampling. In between is a setting which is called Adaptive Multi-Sample AA and is the mode where the driver decides.

So basically it could just be a copy and paste bug from older xbitlabs reviews.

do you mean copy/paste bug as in, they're using the middle mode but say quality, or that it's just that the explanation is copy/paste "bug"?

 

[CarstenS]

IMHLO it could either mean:
• In sovjet russia, the CCCP chooses your anti-aliasing setting!
• they just copied the whole testbed-section from an older review and didn't use any transparency- or supersampling enhancements
• they're still using the middle mode and just dub it like they were used to.
• they're really comparing the most "quality" setting, i.e. sparse grid supersampling vs. transparency multisampling.

#2 and 3 I find most probable.

 

[Alexko]

IMHLO it could either mean:
• In sovjet russia, the CCCP chooses your anti-aliasing setting!
• they just copied the whole testbed-section from an older review and didn't use any transparency- or supersampling enhancements
• they're still using the middle mode and just dub it like they were used to.
• they're really comparing the most "quality" setting, i.e. sparse grid supersampling vs. transparency multisampling.

#2 and 3 I find most probable.

No, it's definitely #1!

 

[stevem]

It's not inadvertent or errant pedantry. It's been pointed out & ignored numerous times. I believe they have a methods/rationale page somewhere.

 

[chavvdarrr]

• they're really comparing the most "quality" setting, i.e. sparse grid supersampling vs. transparency multisampling.

Thats what they do usually.
If these modes give ~equal quality comparing is ok.
After all, you either search for settings that give almost equal fps (hardocp style), or compare speed with equal quality.
Equal settings with non-equal picture quality looks like worst scenario.

 

[fellix]

GF106:

GeForce GTS 450

 

[Alexko]

"The GF106 appears to be small, with a side length of only 15.2-15.7mm."

So 231,04 to 246,49mm². I'd hardly call that small, it's almost as big as RV770.

 

[fellix]

More like G92b?

 

[rpg.314]

So 231,04 to 246,49mm². I'd hardly call that small, it's almost as big as RV770.

So if GF108 is half of gf106, that will make it ~125mm2. So nothing below $99 from nv now?

 

[fellix]

GF106 is definitely far away from 1/2 GF104 in die size, so does that confirms the 192-bit interface?

 

[Sound_Card]

I could be wrong, but I thought I remembered that AMD took SS out of adaptive mode and MSAA/SS is decided by the driver per application. They were kind of toying with this if you recall way back when they introduced the Adaptive AA slider modes, and then randomly took them right back out.

 

[Blazkowicz]

so is it another chip with two geometry cluster things? something huge like 288SP?

GF108 could then be a third of GF106 (96SP, 64bit bus)