You are on page 1of 2

Civilization 5 Benchmark modes.

The benchmark modes in Civilization5 are designed to stress and test various aspects of the users
hardware and supporting software ( e.g. drivers ). There are 3 benchmarks in total. Each benchmark
tests a specific workload scenario. In addition, any of the tests can be run with command line modifiers
which alters the way threaded rendering submission is handled by Direct X 11.
To ensure the benchmark results are reported, in the config.ini ( located in your “\Documents\My
Games\Sid Meier's Civilization 5” ) make sure that “LoggingEnabled = 1”. The results will be stored in the
logs folder at the same level as the config.ini. Ensure that “MaxSimultaneousThreads” is set to the
number of processor cores you wish to test against. Values for MaxSimultaneousThreads greater than
16 are unsupported and clamped.

1. Late Game View Benchmark.


This benchmark is designed to simulate a late game workload. This scenario exercises all aspects of the
game engine pipeline since all simulation and renderable object types are represented at a frequency
consistent with a game that has been in progress for 300+ turns. To run this test, run the application
with the command line option “-benchmark lateGameView”. Case insensitive.
Results are reported in the “LateGameViewBench.log” file.
Full Render Score reflects the active rendering settings of you application.
No Shadow render score turns the shadow pass off for the application.
No Render score emulates an infinitely fast GPU and will bypass most driver overhead. This serves as a
baseline for Civ 5’s engine performance vs GPU and Driver performance. ( note while in this mode, the
display will flicker ).
All results are reported as frames rendered over 60 seconds. Additionally the number of draw call made
per frame is reported.
Note: Mousewheel zoom in some benchmarks are enabled for inspection, however using this will affect
the end results. For consistent results the user should avoid altering the zoom level.

Unit Benchmark.
This benchmark is designed to stress test the users system by executing a parallel series of animation
and rendering tasks. The workload will heavily stress the CPU as well as GPU and driver. It is designed as
a total throughput test to evaluate a users system and determine where the bottlenecks occur. Results
reported are similar to the Late Game View benchmark, however the display settings are overridden for
consistency. Therefore changes to the graphics settings will not impact the test except for setting
resolution. To run this test, run the application with the command line argument “-Benchmark Units”,
case insensitive.

Leader Benchmark

The Leader Benchmark is designed to test advanced rendering features which we use for our leader
scenes. This test can run through any of the leader scenes specified and report the performance for that
leader. Also this test can be used to measure the performance of Compute Shading features via our
variable bitrate compression technology. This compression test measures the GPGPU performance of
the video card. To run in leaderhead benchmark mode, start the app with “-LeaderBenchmark “
To run in compression mode, start the app with “-LeaderBenchmark –compression”.
Note: Compression mode will do nothing but continually decompress and display textures for the given
scene for the specified duration. The leaders themselves will not be displayed.
The full syntax for the command line options are:
-LeaderBenchmark [-compression] [-duration time_in_secs] [-norendering] [-logname log_filename]
<list_of_leader_files>
-logname is the name of a file to append to (Default is 'LeaderBenchmark.csv')
-duration is how long to remain in each scene
-norendering is equivalent to the norender option in the previous benchmarks.
<list_of_leader_files> is a list of XML files (e.g. Catherine_Scene.xml OdaNobunaga_Scene.xml)

The benchmark dumps a file into the app directory named 'LeaderBenchmark.csv' which contains the
resulting information. Statistics are given for each leader independently, and they're appended onto the
file from one run to the next. Mean FPS is the statistic that matters. It's an average over the entire run
for each leader, with the first 30 frames ignored to prevent load time and startup hitches from skewing
the results.

General rendering options.


These options allow advanced users to change the way Civilization 5 handles threaded display lists (DL)
on DX11. For more information about this, users are encouraged to read the DX documentation. Note:
this is for advanced testing for threaded GPU drivers. These settings are also stored in the config.ini file.

“ThreadingMode = 0”
Controls the threading strategy: (0=default,1=no display lists,2=one DL per command set, 3=split mode,
4=aggregate mode).
0 - Do whatever we think is best (equivalent to 2currently)
1 - Disable display list use and do everything on the immediate context
2 - Our current behavior
3 - Split and merge command sets, aiming for a particular display list size.
* This mode is aiming at a balanced load. It will squish together small DL's, but will also rip large
ones apart. This may produce more DLs than our current path does, depending on the workload.
4 - Combine small command sets into one display list, but never split them.
* Should be as good as 'split' at eliminating small lists, but won't try to balance the load.

“TargetJobSize = 100”
Number of commands per diplay list to aim for in SPLIT and AGGREGATE thread modes.

You might also like