Ghost Recon: Advanced Warfighter 2
The first game we decided to test was GRAW2, a title notable for its PhysX implementation. The game features an extra level accessible only to people who have a PhysX accelerator installed. The level, aptly named AGEIA Island, features fully destructible environments with huts and fences being fair play.
Because the performance levels on this map do not exactly match the performance in the campaign and because we were unable to test the map without acceleration enabled (PhysX acceleration is hardcoded to enabled in this level) we also tested the performance in the single player campaign, benchmarking spots where physics calculations were being performed due to in game events such as explosions.
As you can see from the above graphs, enabling PhysX acceleration actually dropped the performance. This is due to the fact that with acceleration enabled the game renders far more particle effects when things explode, resulting in a considerably higher workload for the graphics card. The same is true even if detail levels are set as low as possible.
Looking at the CPU load reveals an entirely different picture. The minimum load was more or less the same with acceleration enabled or disabled. This comes as no surprise, since the value basically represents the load on the CPU during the times when no physics calculations were required such as when moving around. The maximum value differs greatly at 13%. Comparing the CPU load graph against our gameplay session we can see that the peaks always happened when huge explosions were present. This is where the PhysX accelerator helped things considerably, since the load raised only marginally thanks to its help, whereas with no acceleration the CPU load increased considerably. The resulting average load was higher as well, in part due to the slightly higher load at all times caused by bodies collapsing, bumping into boxes and other physics calculations.
