I don't think SLI is broken there because I get 30fps avg. with a single GTX 460 at 1920x1200 (all maxed with ambient occlusion) so those averages with GTX 460 SLI are about right aren't they?
^Actually yeah thinking about it further, with one card dedicated to PhysX he/she was still getting 48fps. Unless the PhysX effects are a huge bottleneck, it should be faster. Seems like I'm getting very low then with my single 460 at 1920x1200. I do enable TRAA, 16xAF, and ambient occlusion though.
well you will have to actually give us some info as we have no idea about your setup. please list your specs and the drivers that you are using.Dam tried to play the demo, but all I see are clouds and no floor....cut scenes are fine, but in actually game play no chrs or walls..everything is see through in demo.. any Ideas whats wrong?
well you will have to actually give us some info as we have no idea about your setup. please list your specs and the drivers that you are using.
This game obviously uses a heavy amount of gpu-physx when on high settings, so a dedicated physx card *SHOULD* be faster than sli.
This game is going to suck - so why waste time benchmarking?
Interesting factoid here. If you own a non physx card or a single gpu lower than a 460 and want to use physx you can go into the physx folder and delete (back it up first) the folder called CLOTH in the apex folder. Then you can set physx to medium. Yeah it will run on the cpu, but very playable on my 5850. There is also a way you can do it so that you get vitos cloth movements and it still runs okay. Haven't tried that myself.
http://physxinfo.com/news/3628/mafia-ii-demo-tweaking-physx-performance/
That logic doesn't make sense.
As soon as you dedicate one card to PhysX, you immediately limit your graphics performance to a single card.
If you don't dedicate a card to PhysX, then the load can be spread between PhysX and graphics (I believe that PhysX always runs on one card only, and graphics on both).
So unless PhysX maxes out the entire videocard anyway (which I don't think it does with a high-end GeForce), using SLI would give you more graphics performance than a dedicated PhysX card and a dedicated graphics card.
Ofcourse, in practice problems with the game/driver code could still make SLI perform worse... but in theory SLI should always be as good or better than a dedicated PhysX card.
Do you have a source to confirm that the physx load is still burdened onto just one card when running in sli? I was always under the impression that both cards would be processing physx calculations, which would tie up the shaders and kill the fillrate.
If I had a source I would have given it. But I think the bottom line is that Cuda itself doesn't 'do' SLI. Running Cuda on multiple GPUs is normally done by programming each GPU individually, not via SLI.
And there's a logical problem here:
"tie up the shaders and kill the fillrate."
For example:
1) Shaders aren't necessarily tied up all the time, there's still quite a bit of idle time during graphics. This time can be used by other workloads.
2) There's more to fillrate than just shader power.
Ok it looks like you're right. And even more interesting, it appears that in an SLI setup, specifically having the second card perform the physx operations results in better performance.
http://www.overclock.net/nvidia/786747-brettjvs-batman-physx-benchies-interesting-results.html