Fractal flames require lengthy renders because of the nature of the algorithm. It is both computationally intensive, and memory cache inefficient. This holds true for running on both the CPU and GPU.
Many benchmarking tools focus on either computation or memory bandwidth. Few do both, and even less measure memory latency. An algorithm that heavily stresses computation and memory latency makes for a good real-world benchmark.
This is accomplished by running the EmberBench.ps1 Powershell script in the Bench subfolder of the installation directory on Windows.
For benchmarking on Linux, you must first install Powershell, then run pwsh
, then run EmberBench.ps1
out of /usr/share/fractorium/bench
following these steps:
- You must enter the folder as root to successfully run it, so navigate to
/usr/share
and right click on thefractorium
folder and select Open as root. - Before running the script, open it in an editor and see the instructions for editing the device index of the GPU. This varies by system, so you might need to run
emberrender --openclinfo
to determine your device index. Note you can also benchmark with multiple GPUs if you have them, by specifying the device indices as a comma separated list. - Save and close the file.
- Run
pwsh
then invoke the script by running./EmberBench.ps1
After the script completes running in about 30 minutes, the output will be in the parent folder in a file named benchout.csv which will show the iterations per second performed for each test flame in single and double precision modes, with various supersample values, each ran separately on the CPU and GPU.
It is the aim of this project to get fractal flames rendering more accepted in mainstream benchmarking.