GPU PhysX: Roundup: Free For All
Test Sequence and Combinations
We start by combining our test subjects and benchmarking them in the following configurations:
- AMD main graphics card + GPU PhysX (Nvidia card)
- Nvidia main graphics card + GPU PhysX (Nvidia card)
- A single graphics card running GPU-based PhysX
- CPU-based PhysX
Instead of using the games Metro 2033 and Cryostasis for benchmarks, we opted for the recently-published Mafia II. Its ratio of graphics to physics is quite balanced, and it allows us to make a direct reference to a current game so our recommendations are more relevant.
|OS||Windows 7 Ultimate x64|
|Game||Mafia 2 via Steam|
Below is the chart we created using the different combinations of graphics cards and manufacturers:
As expected, using a dedicated graphics card for PhysX makes a difference. Pairing it with a high-end model from each camp results in a rather even playfield. The GeForce GTX 480 can neither pull ahead much from the Radeon HD 5870, nor really make the GeForce GTX 460 and Radeon HD 5850 eat its dust. All of the GPU + GPU combinations are significantly faster than using just a single Nvidia card for both graphics and PhysX.
The single cards are already dangerously close to the lower limits of playability. The chart shows the average frame rates, but obviously the difference will be seen most clearly in minimum frame rate numbers. Most of the time you will be walking around, and the frame rates will be the same regardless of whether you are using a dedicated PhysX card or not. But as soon as something happens that requires physics calculations, that's where the difference lies. Since this happens only briefly and occasionally, we chose to show you the overall picture instead.