Crysis Benchmark Results

Crysis uses a new graphics engine: the CryENGINE2, which is the successor to Far Cry's CryENGINE. CryENGINE2 is among the first engines to use the Direct3D 10 (DirectX10) framework of Windows Vista, but can also run using DirectX9, both on Vista and Windows XP.

Roy Taylor, Vice President of Content Relations at NVIDIA, has spoken on the subject of the engine's complexity, stating that Crysis has over a million lines of code, 1GB of texture data, and 85,000 shaders. To get the most out of modern multicore processor architectures, CPU intensive subsystems of CryENGINE 2 such as physics, networking and sound, have been re-written to support multi-threading.

Crysis offers an in-game benchmark tool, which is similar to World in Conflict. This short test does place some high amounts of stress on a graphics card, since there are so many landscape features rendered. For benchmarking purposes, Crysis can mean trouble as it places a high demand on both GPU and CPU resources.

Low-resolution testing allows the graphics processor to plateau maximum output performance, which thereby shifts demand onto the system components. Even still, Crysis appears to have a preference for the new PCI Express 2.0 graphical interface found on G92 graphic processors; even if it only hints at it in our results.

Even without Anti-Aliasing turned on, Crysis keeps the top four competitors around 60 FPS. It's clear that the CryENGINE2 is a heavy hitter, as the insanely overclocked ZOTAC 8800 GT AMP! Edition outperforms the GeForce 8800 GTS 640MB by over 26% and the 8800 GTX by almost 3%. The most surprising result of all is the 9800 GTX, which outperformed everything else at this resolution and leads the 8800 GTX by nearly 11%, 9800 GTX by 6%, and the 8800 GT by 8%.

Even with more vRAM available to them, the older 8800 GTS and GTX just cannot offer the performance of the G92 GPU paired with the PCI Express 2.0 graphics bus. What comes as a surprise to me is how the single G92 9800 GTX was able to out-perform the two G92 GPU's inside the GeForce 9800 GX2, which did not shine so bright in this low-resolution test.

At the 1280x1024 resolution which mirrors what gamers using a 19" LCD monitor would experience, the results are still excellent but also beginning to near the 30 FPS threshold for acceptable performance. The aging G80-based GeForce 8800 GTS unit takes a pretty heavy hit, and drops clear out of the desired range.

In terms of general performance, all of these products maintain the same performance ratio as before. This gives the 9800 GTX a 10% frame rate improvement over the 8800 GTX, but nowhere near the performance of the 9800 GX2 which is beginning to pull away from the pack with more than a 22% lead.

Surprisingly, the three GeForce 8800 series products maintained a rather constant performance ratio between one-another throughout the Crysis benchmark tests. While the ZOTAC GeForce 9800 narrowly outperformed the G80-based GTX by only one frame per second on average, the GeForce 9800 GX2 actually improved as the demand increased resulting in a 35% gain. But in the end, these were all tests with no post processing effects; and who buys a high-end GPU to play PC video games without AA anyway?

Crysis offers a very wide range of settings beyond the basic "High" setting, which allows Anti-Aliasing steps of: No AA, 2x AA, 4x AA, 8x AA, 8x Q AA, 16x AA, and finally 16x Q AA. At the end of our real-world testing Crysis was given the GPU-thrashing 16x Q AA performance setting, and we watched the G92 youngsters run circles around the aging G80 generation. Most enthusiasts have lost sight of the fact that the G80 GPU doesn't offer the post processing memory optimizations and compression available to the new G92 series. This functionality makes all the difference when AA is turn up.

Thanks to the AA optimizations in the G92, ZOTAC's GeForce 8800 GT AMP! Edition outperformed both G80 GPU's by a very significant margin despite all of the previous tests indicating a much closer disparity. It's obvious that the new 65nm G92 core architecture is to be credited, and perhaps the new PCI Express 2.0 interface which allows twice as much graphics data bandwidth also played its part. Then again, maybe MSI's factory overclocked 8800 GTX isn't built to keep pace when the post processing is maxed out. Either way, our benchmarks below certainly indicate that while the 8800 GTX matched the 9800 GTX in other tests without AA, but it doesn't even come close in a high-pressure Crysis.

With an incredible performance difference of 40% launching the 9800 GTX over the 8800 GTX, the difference between generations becomes clear. Even ZOTAC's 8800 GT AMP! Edition video card boast an 37% improvement over the more expensive 8800 GTX. Ultimate praise is due towards the unfathomable performance exerted from the 9800 GX2, which snatched a 38% over the G92 GTX.

While the extra load did show an impact on the performance results, the all of the G92 video card walked over the competition with uncanny difference between them and the G80 versions. If you want to play Crysis will bells, whistles, and bag pipes, and you're not in a position to use an SLI array, then the 9800 GX2 is the obvious choice; but that ZOTAC 8800 GT AMP! Edition video card wins the price-to-performance nomination hands-down.

EDITORS NOTE: After many months of using the Crysis demo for testing with the MadBoris Benchmark Tool, we recently started using the full retail version. Our initial tests have discovered that non-AA results were identical, but the 16x Q AA test produced very different test results. All of our previous results are still good for product comparison, but using the patched retail version of the video game (v1.21) has demonstrated that post processing effects (offered up to 16x Q AA) were not fully incorporated into the demo (limited to 8x AA). We have decided to address this matter in our Crysis performance comparison: demo vs retail forum discussion and welcome your comments.

In our next section, Benchmark Reviews switches to video-output only benchmarking, and uses Lightsmark for an apples-to-apples comparison of performance.

<< Prev - Next >>