IHV-specific Enhancements in Games

TressFX use DirectCompute, no ?

So it should work on nVidia hardware too... ?

 

Yeah, but you'll see the same thing as Dirt:Showdown where a gtx680 can barely match a 7950.

 

Even after some drivers revisions and optimisation time ?

 

I think enough time and driver optimizations/revisions have passed since Dirt:Showdown was released.
http://www.computerbase.de/artikel/grafikkarten/2013/test-nvidia-geforce-gtx-titan/35/

 

Considering how long nVidia banged on about their compute advantage while it was mostly useless, they can hardly complain when they decided to sacrifice that performance right when it was becoming practical.

 

Yes, even then. 7970 GE even beats Titan.

 

Yup if Nvidia cards are going to struggle due to a weakness then they'll just have to improve performance much like AMD had to with tessellation.

 

OlegSH, your attitude towards this is quite curious, considering it's not AMD of the 2 companies which has the history of screwing users of other brand over and over again.

To my knowledge, only Forward+ lighting is something nVidia still hasn't been able to get running at proper speeds.

 

The problem with these high tessellation levels is that the miniscule IQ gain (some would say it is actually a loss as it looks less realistic in some cases) is nowhere near worth the performance loss on both AMD's or Nvidia's cards. However, AMD cards are generally impacted harder so it's a net gain for Nvidia to hobble their own cards with artificially high levels of tessellation.

The difference in this case is clear - there is a very obvious IQ gain with TressFX. Whether or not that's worth whatever performance hit there is will be down to the player to decide.

 

Haven't really seen a nvidia demo with Forward+, but all forward+ research done on nvidia hardware seems to work just fine..

 

Kepler has displayed pretty poor compute performance in a number of compute benchmarks, often falling behind even Fermi. Besides, Dirt Showdown came out just a couple a months after GK104 came out, which means there's no way AMD could have foreseen that it would suck at compute when they started cooperating with Codemasters, especially since the Foward Lighting technique used was first displayed in the Leo demo that came out even before GK104.

And AMD never pushed for proprietary libraries (PhysX) with a sabotaged CPU codepath, never removed objects altogether or disabled AA from games when an AMD card was detected (Batman AA), never tessellated flat or non rendered surfaces (Crysis 2), sabotaging performance for both AMD and NVIDIA users (just to a larger extent for AMD users), etc.

They just don't behave the same way when it comes to this kind of stuff.

I wouldn't say that. Aliens vs. Predator made very good use of tessellation, for example. I don't remember any instance of tessellation making things look less realistic, except maybe the cobblestones in Unigine Heaven, but that was a demo meant to showcase tessellation, so it was understandably pushed to the extreme.

 

Was referring to the Dirt Showdown case where advanced lighting is Forward+ and kills performance (relatively) on the GF's.

Oleg,
GTX680 for example loses to GTX580 in many compute cases, I wouldn't call that "plenty fast"

I assume you do have some proof of "AMD using slowest versions because they're slower on nVidia" or back engineering them to be purposely slow?

 

It's enough to play to one's own strengths, just as Nvidia did with overtessellation.

 

They might not be written to run efficiently on NVIDIA cards, but per pixel linked lists are IMO going to see a lot of use going forward.