Hawken PhysX Analysis



Company: Meteor Entertainment
Author: Sean Ridgeley
Editor: Charles Oliver
Date: January 25th, 2013

Frame time test, 99th/99.9th percentile test, conclusion

Frame time test

As mentioned, framerate reports alone only tell part of the story with the performance of any given game. To offer a deeper insight into how a game performs, testing frame time how long it takes for the GPU to deliver each frame is critical. To this end, FRAFS Bench Viewer software is used in conjunction with FRAPS for this test and the one following.

What you want in a graph like this is an end result that's mostly even and short in height across the board, thus indicating tight, consistent performance. Spikes are expected and do little to noticeably impact the player experience if they occur rarely. Long, chunky, block-shaped sections, however, depict major stuttering and uneven performance that does noticeably hinder the experience.

The graph shows with PhysX on, Hawken frames are regularly delivered at a respectable 8ms or below, but about as often, they're not. More, frames delivered above the 8ms average become extremely inconsistent, with frequent and wild spikes that sometimes reach about 26ms, and generally erratic behaviour even besides these. Fortunately, spikes are often only noticeable in games at 60ms or beyond, but particularly sensitive players may find the inconsistency here disconcerting during gameplay. These results are in contrast to those with PhysX turned off, which are significantly more consistent in testing, though not much noticeable in practice.

99th / 99.9th percentile frame time test

Exploring frame time further, the 99th percentile test depicts the figure (in ms) at which 99% of frames were rendered. What you're looking for here is a consistent line that deviates only at the last 1% section of the graph, indicating consistent, low-latency frames.

The graph shows results in line with the frame time tests, which is to say performance isn't drastically different from PhysX off tests, but different enough to be noteworthy. The 99th percentile frame time figure is still at about 11, which says 99% of all frames were rendered at that figure or below, which is to say quickly.

The 99.9th percentile frame, meanwhile, depicts the figure (in ms) at which all but 0.1% of frames were rendered faster than. Because the 99 and 99.9% figures are very far apart with respect to minimum frames per second, it can be concluded stuttering is much more present in Hawken with PhysX enabled, though still not necessarily to the point of being distracting or even noticeable (in my testing, there wasn't much to speak of).

Conclusion

Hawken is reasonably well-optimized for PhysX, though testing shows it isn't perfect. Nevertheless, if you can spare the headroom, it's certainly a recommendable option that offers appreciable visual differences; if you can't or if you're running AMD, know the differences, while nice to have, are more or less subtle.