Crysis Benchmarking - Methodology and Mythbusting

[Nedjo]

For a long time I was trying to find out appropriate way to benchmark Crysis.

I wasn't satisfied with what mainstream media was doing with pointless flyby demo for GPU, and even worst demo for CPU testing.

I've done some digging and will present you in this thread what I've find and what I think is appropriate way to benchmark Crysis.

I hope that you'll find this useful, and will accept this as a default way for testing Crysis, so that we all can share and compare our results.

First of all, what's needed is a good front end app that will allow easy changes for settings. Luckely there is outstanding little app for this purpose!

Second what we need is appropriate benchmarking sequences. To my amazement (maybe some of you knew that) Crysis have benchmarking sequence for each single level!

When we mix up front end app, and do appropriate editing to have those sequences usable for benchmarking we get this archive that you need to download:

http://rapidshare.de/files/38479006/...rking.rar.html

Just unRAR this archive to the folder where you've installed Crysis (e.g" E:\Games\)

You can run benchmark for any level in Crysis, by simply choosing appropriate demo recording from level folders. You can even run demos for multiplayer levels.

In each folder there are two timedemo files. You need to select file with the same name as the level.

I've tried all timedemos, and it seems to me that demo from level "sphere" is the best one! There is a plenty of jungle, water, snow, ice, shaders… really the best representation of the majority of game.

You can see my results for "sphere" (Vista64, Phenom 9700, Radeon HD3870, Catalyst 8.1)


I usually run timedemo three times. You'll get average fps from all three runs at the end. But bear in mind that this nice app doesn't place coma at the right place in the resulting score. If you want score for each run, just scroll up slider…

You can setup bunch of different settings, so feel free to experiment!