Which platform is the fastest with Huygens?

Comparison of deconvolution speed with Huygens on various platforms (Linux, Mac and Windows).

Clearly, the type of processor and its clock speed are very important for system performance. Still, next to the processor and the number of cores, there are many factors affecting system performance: type of memory, chip set which interfaces cpu to memory, speed of cpu interconnect, cache size, number of caches, and also software issues like the Operating System (OS) the system is running on.

Windows vs Linux vs Mac OSX

In recent cases where Windows and Ubuntu were tested on the same hardware Ubuntu Linux outperformed Windows 7 by about 30%, in one case even close to 40%. We know of just one test in which Windows was run on Apple hardware. In that test Windows outperformed Mac OSX by 5%.

64 bit vs 32 bit

64 bit applications require 2 times more storage for address and size information, which might cause a slow down due to increased memory traffic. For example, on Irix 32 bit applications tend to be 5-10% faster than 64 bit applications. Still, for many of today's microscopic datasets, the memory limitations imposed by 32bit processing are far too restrictive to consider using it for a small performance gain. For this reason in the tables below you'll find mostly tests of 64 bit binaries, the exception being MacOSX older than 10.6 which has limitations in running 64 bit applications.

Weighted score

The weighted score is a measure of the performance of a machine, relative to a dual AMD Opteron 2.2 GHz computer (which has a weighted score of 1). We achieved the following Huygens Benchmark figures for the speed of deconvolution:

These figures might work out differently for different image sizes, or different thread counts.