Man of Honour
That's odd, given that PhysX is designed to be run through a GPU I find it strange that a CPU handles it better from your experience. Is this only the case with the extreme Intel CPUs (hexcores) or is it also similar with the slightly lower performance ones like the i7-4770K and i7-3770K etc?
I am guessing it is because of the overclocked hexcores. I have just been checking old benchmarks and @1080p the stock Titan plus dedicated PhysX card is slightly faster than the 290X but @1600p it is the other way round with the stock 290X beating the Titan plus dedicated PhysX card. I think this is because of the lower fps @1600p which gives the CPU less work to do.
http://forums.overclockers.co.uk/showpost.php?p=25193535&postcount=50
In the link above there is a stock Titan plus another Titan doing the PhysX
@1080p
Min = 71
Max = 132
Ave = 94
@1600p
Min = 42
Max = 74
Ave = 53
http://forums.overclockers.co.uk/showpost.php?p=25232838&postcount=101
In the link above is a stock 290X @1600p using the same settings as the Titan above
@1600p
Min = 29
Max = 79
Ave = 56
http://forums.overclockers.co.uk/showpost.php?p=25232639&postcount=99
In the link above is a stock 290X @1080p using the same settings as the Titan above
@1080p
Min = 30
Max = 135
Ave = 88
You can see several things from the numbers above
@1080p the Titan is faster
@1600p the 290X is faster
The minimums for the 290X is lower at both 1080p and 1600p, I have put this down to the CPU also doing the PhysX.
I think if the 290X was to have a dedicated PhysX card the same as the Titan above has, the 290Xs minimums would come up and it would win at both 1080p and 1600p by a wide margin.