Benchmarks: MSAAx8, Physx Normal
With MSAAx8 enabled and PhysX set to normal, which offloads physics calculations to the CPU if you're on an AMD card, while Nvidia's can handle either GPU or CPU-powered PhysX so we ran the GeForce cards in both configurations.
At 1680x1050, the GTX Titan using GPU-accelerated PhysX was the fastest graphics card tested at 96fps while the R9 290X trailed at 86fps, which was slightly faster than the GTX Titan using CPU-calculated PhysX. The GTX 770 was good for 69fps with GPU-accelerated PhysX, while this figure dropped to 61fps after switching to the GPU, making it just 1fps faster than the R9 280X.
At 1920x1200, we see that the GTX Titan still holds a substantial edge over the R9 290X, averaging 86fps with GPU PhysX and 76fps without, whereas AMD's flagship produced a slower but perfectly respectable 73fps. The R9 270X took the bottom of our chart here, but it did well to average 43fps, 7fps slower than the GTX 770 when using CPU PhysX.
The GTX Titan placed as expected at 2560x1600 with 53fps (GPU PhysX), while it only bested the R9 290X by one frame when both relied on CPU-accelerated PhysX.