Crysis

First person shoot-em-ups are probably some of the most stressful applications you can run on any PC. For a number of years I have used time demo benchmarks from Quake and UT and a variety of other games to test CPU, video card and memory performance. Unfortunately it got to the stage where video card manufacturers started to optimize their drivers for benchmarks, doing horrid cheats like lowering graphic settings during benchmarking to make their product look fast in benchmarks. So now in, most cases, running simple time demo benchmarks is only viable when tweaking your own hardware. Since the only thing we're changing in this setup is the CPU, benchmark optimizations shouldn't be a factor for concern.

For the first person shoot-em-up section of these tests, I decided to use Crysis, since it was the only fps game that I had on my shelf that actually taxes my hardware. I played through the entire game and was disappointed when I was nearing the end of the game, because particular sections of the last fights actually turned into a slideshow and I had to go back and lower the graphics settings on my mid-range PC to get a smooth but still sexy looking gaming experience.

I have tested crysis in a couple of different ways. First I tested the default built-in benchmark batch files, then I also tested two ingame areas using fraps.

Official Benchmark

During the first CPU test, the quad core offered only a fractional performance increase.

Crysis CPU1 Test, dual core results

Even though we've seen little to no performance increase in the benchmarks, the CPU utilization graphs show a very interesting story with the dual core CPU in this first benchmark. The CPU is doing a lot of work on both cores, almost capping out CPU utilization in several parts of the benchmark.

Crysis CPU1 Test, quad core results

The quad core on the other hand had medium utilization across three of the four cores and low CPU utilization on the third core as you can see in this graph. Again, like in World in Conflict, although we are not seeing any real performance difference, the quad core demonstrates that it still has a lot of headroom.

CPU2 32 bit Test

Another thing I wanted to analyze was the way in which the CPUs were treated when running applications in 64bit rather than 32bit. For this reason I chose to run all my tests on XP 64bit and Crysis offers the ability to benchmark in either 32bit or 64bit mode. This next benchmark was run in 32bit mode. Rather than bore you with the 64bit identical benchmark, we'll just show you the figures from the 32bit version. During my tests I found no performance difference between 32 and 64bit versions of Crysis's benchmarks.

The CPU2 benchmark includes particles in the air in the form of snow. You can really see the stress difference on the CPUs during these tests. The quad core CPU dropped as low as 8.64fps and the poor old dual core didn't even crack the 1fps at one stage. This is one scary test for CPUs!

Crysis CPU2 Test, dual core results

The dual core sat at 90% CPU utilization for most of the test. As you can see, both cores were heavily utilized.

Crysis CPU2 Test, quad core results

On the quad core side of things, only one of the four cores looked like it was doing really hard work, the other cores were kind of cruising with lots of headroom. I'd say going to an SLi solution, or even upgrading to the next generation graphics cards, you'd see this CPU able to truly shine in this benchmark. Although we don't see massive performance increases in the average scores, we do see the quad core not working anywhere near as hard, again.

Playtests

The first play area I tested was a scripted movement section of the game which involves the player sitting in the back of a Humvee, shooting and attacking Aliens. I was able to replicate almost identical movements during gameplay.

The quad core just edges out the CPU in performance. 4 fps isn't a great deal at the maximum level but the fact that the quad core didn't drop below 30 is kind of important. I remember someone telling me that the eye can't see anything above 30fps. A standard TV in the US (with the NTSC standard) streams at 29.97 (approx) fps. So I always find when playing fps games, it not a matter of how fast the fps are, it's a bout how low they go during hardcore scenes. In all honesty it doesn't make much difference if your video card streams along at 180 or 80fps, both are seen as perfectly smooth gameplay. I would say that both the CPUs in these tests were more than capable of doing everything required, the only limitation was the graphics card which unfortunately was the fastest single card solution available on the market at the time of testing.

Looking at the above graph, you can see the performance of the two CPUs during 60 seconds of the test. Both performed almost identically and CPU utilization was very similar to what e saw in the CPU1 benchmark.

Moving on to the end boss, the experience was the same as we saw with the Humvee, with a massive dip in frame rates as a massive explosion occurs during gameplay. Both CPUs dropped down to 12/13 fps and it was a noticeable jerk in gameplay. The quad core CPU demonstrated much faster high end frame rates, peaking at almost 10fps higher than its dual core counterpart. At that level, it's a 20% performance increase.

The benchmark was started in the same little doorway area, just before you walk out onto the main deck of the battleship in an enclosed area. It's in this enclosed area that the quad core screamed along at over 50fps. Once I was out on the deck, both CPUs demonstrated very similar performance. When you look at this graph you will see the lows and highs in different positions - this is only due to human interaction with the gameplay - although I made all the same moves, I may have made them at very slightly different times during the sixty seconds.