Search Prices:

FarCry

There are several demo options FarCry Benchmark; we settled with “River”, "Pier", and “Volcano”. The "River" map features an enormous waterway and wall-to-wall foliage, frequently exceeding 1.5 million polygons per frame. As such, this demo presents one of the more difficult rendering loads. The "Pier" begins with a similar scenery setup, but the complexity varies widely; the polygon count jumps as high as 1.5 million but primarily hovers between 300,000 and 750,000 and briefly flirts with 10,000. On average, this provides an intermediate rendering load. The "Volcano" map is set inside the barren volcanic bowl, with an overhead walkway as the sole scenery; the polygon count stays below 75,000 per frame for the duration. This demo presents one of the lightest rendering loads.

We selected maximum graphics settings. Geometry instancing, normal maps compression, and HDR were enabled. FarCry allows HDR settings from 1 to 11, with 1 being barely noticeable and 11 mimicking the iris effect of a $50 digital camera overdue for an exorcism. Our formal tests used an arbitrary setting of 9 after experimentation indicated that other settings did not produce a discernable change in results. We had previously set the console variable that uses all-3D vegetation, rather than drawing sprites beyond a certain view distance, and found that this setting was maintained during the benchmark.

Three passes were run, with the utility configured to take an average of two consecutive runs at each resolution. For the first pass, we set anisotropic filtering (AF) and antialiasing (AA) to 1 and 0, respectively. Then we stepped to 4+4 and 8+6 to see how performance was affected.

For the River demo, the results looked like this:

Frames per second: Higher is better. The change in average frame rates is dominated by changes in resolution. The legend shows Anisotopic Filtering (AF) + AntiAliasing (AA), in that order.

Essentially the same results were found in the "Pier" level.

The three AF/AA settings basically track each other for the first three resolution settings, although at 1024x768, there are hints that the additional load is starting to impact performance. At 1280x1024, we find a noticeable performance dip from the resolution alone, and nearly the same penalty again if AF/AA settings are enabled – although interestingly, the degree of AF and AA does not seem to make a difference in the results; ‘on’ or ‘off’ is determinant.

We were curious if the level of anistropic filtering and/or antialiasing was more to blame for the droop, and attempted a series of tests at 1280x1024 in which AF was varied while AA was disabled, and then vice-versa. The results suggested that each was contributing about 50 percent, and confirmed that the degree of AF or AA made no difference.

Then we ran the Volcano demo:

Frames per second: Higher is better. The change in results is once more dominated by changes in resolution. Note that the X and Y axes have been switched for a different perspective, that is, to show the impact of AF and AA. It is apparent that just using the 4+4 setting causes approximately a 20 % penalty, however, going to higher AA+AF levels goes rather unnoticed with respect to performance.

With the typical polygon count reduced by a factor of about 20, the framerates of the two lesser resolutions fly for the stars, and even at 1280x1024, the unfiltered framerate is close to the magical 60fps mark. Once again, AF/AA settings inflict an ‘on’ penalty with no respect to degree, although the drop starts to become noticeable at 800x600 this time around, and at 1280x1024 the difference has widened to almost 10fps.

These tests would suggest that AF and AA are a free lunch if the initial framerate loss can be tolerated. At 1024x768, that criterion seems to be met. To find out if this result would hold true with a more recent game, we switched to HalfLife2.

Next Page:    => Performance: HalfLife2 =>