FERMI Architecture Analysis

[BFG10K]

I don't think tessellation alone is the cause of Metro2033's slowdowns. It does incur a performance hit turning tessellation on, but it isn't game breaking. There are other factors at work with that game which contribute heavily to low frame rates.

56.3 FPS vs 30.6 FPS at 2560x1600 by simply toggling tessellation from “off” to “on”:

http://www.xbitlabs.com/articles/video/display/hardware-tesselation_8.html#sect0

That’s the difference between a playable framerate and an unplayable slideshow, so I’d certainly call that game breaking.

Civilization 5 does not slide show with tessellation

Has it been confirmed that the “low” setting actually means “off”?

and with what yesterday's HAWX 2 benchmark demonstrated, there will not be any significant slow downs either with this game (releasing in two weeks).

Let’s see how visible the implementation will be in-game. If it’s like Stalker or AvP 3 then it’s worthless.

 

[Lonbjerg]

I have the hardware and the games. What have you got?

A better understanding of the technolgy it seems.
You mentioned "Metro2033".
A game that only tessellate characthers and animals...and a rather clugdy implementation.

And even only bothers to cherrypick a single resolution...

Using that game to gauge DX11 tessellation performance is quite funny....but a waste of time.

So I repeat:
You got nothing?

 

[BFG10K]

A better understanding of the technolgy it seems.

It’s “technology”.

You mentioned "Metro2033".

I actually mentioned several games. Do you read what I write?

And even only bothers to cherrypick a single resolution...

That's the resolution I play the game at.

But if it makes you feel better, the 5870 outruns the GTX480 at 1920x1200 when tessellation is enabled.

Using that game to gauge DX11 tessellation performance is quite funny....but a waste of time.

What should we use instead of actual games? Synthetic benchmarks?

Do you play games on your card, or do you play synthetic benchmarks?

So I repeat:
You got nothing?

You can keep repeating it, but it just means your posts add nothing worthwhile to the discussion.

 

[railven]

I actually played through Metro 2033 this weekend (Steam sale FTW.)

Tess on on my 5870 dropped me from 40-50 in Very High to 20-30.

That to me is a pretty big chunk of performance and frankly I for one couldn't off the bat tell the differences. I'd say the same for DOF (which was even more crippling.)

 

[AtenRa]

Low means NADA, zip, zero Tess in Civ 5

Civ 5 with Tess ON is faster than DX-9 mode with both AMD and NV hardware

HAWX 2 is above 100 fps at 2560x1600 with GTX480 and Tess ON.

 

[thilanliyan]

You realize that this 'emphasis' is merely your subjective interpretation of my words...

Yes that is my opinion like a lot of what it is posted around here. In this case I'm pretty sure other people share the same view as I do about your view of tessellation (judging by the amount of people that have said the same thing as I have).

 

[tviceman]

56.3 FPS vs 30.6 FPS at 2560x1600 by simply toggling tessellation from “off” to “on”:

http://www.xbitlabs.com/articles/video/display/hardware-tesselation_8.html#sect0

That’s the difference between a playable framerate and an unplayable slideshow, so I’d certainly call that game breaking.

You got me there. Looks like the gtx480 has some serious fill rate issues when pushing all of it's resources at that high of a resolution

Has it been confirmed that the “low” setting actually means “off”?

(In regards to Civ 5) Yeah I'm not really sure, but that doesn't change the fact that it does noticeably look better and does NOT run at a slide show.

Let’s see how visible the implementation will be in-game. If it’s like Stalker or AvP 3 then it’s worthless.

There is a benchmark out and it does noticeably look better in-game. I have verified this myself. In fact it was the first time (apart from Unigine's heaven demo which is not representative of an actual game) where I instantly noticed a difference. It still runs amazingly fast on even a gts450 at or below 1080p resolutions.

 

[tviceman]

nevermind!

 

[Scali]

Yes that is my opinion like a lot of what it is posted around here. In this case I'm pretty sure other people share the same view as I do about your view of tessellation (judging by the amount of people that have said the same thing as I have).

And there are also plenty of people who share my view. Doesn't matter really... this topic isn't about views or opinions, but about an analysis of tessellation performance on the Radeon and GeForce architectures.