MC Benchmarks

Benchmarks measure how the code performs as a function of computer hardware as it applies to the EPR fitting problem.

(These results cannot be compared directly with other benchmarks but can aid when considering new hardware)

All results are welcome, so please send them to me!

INSTRUCTIONS: To do benchmarks, install and run the MultiComponent program, go to the "About" tab and press the button labeled "benchmark this computer". Sit back and relax. By default, it will do 9 test sets (serial, parallel, cached) and then pick the fastest each. After the process has completed, press "copy to clipboard" and paste it (ctrl+v) into the body of a new e-mail message. Send it to me for inclusion in the table. Thanks! (try "my last name" at gmail.com)

The benchmark code has been completely rewritten from scratch and optimized for the testing of single core performance, hyper-threading, parallelization, and caching. The results are directly comparable and are in units of Hz (spectra/second). A spectrum is either a plain spectrum or a single MOMD orientation. Since all MOMD orientations can be calculated in parallel, a computer with a high core count is a tremendous advantage.

>>>> Click here to see the Benchmark results <<<<

Some notes about benchmarking:

    • Many factors are important, such as CPU architecture, cache levels and sizes, clock frequency, hyper-threading, RAM speed and amount and number of cores.

    • By default, the program will measure each metric 9 times, then pick the fastest. This assumes that all external influences tend to cause delays.

Here is a sample result showing the serial and parallel performance for a few select systems. Note that the 16(32) core system is over 17x faster when allowed to use all cores. A more detailed discussion can be found here. As of January 2015, the new record holder is a Quad Xeon.

Note that in typical use, only the parallel performance is important!

Units are Hz (spectra/second)

Older versions

Over the evolution of the program, the benchmarking code has received significant changes, and will continue to do so as needs arise.

MC Benchmark V0.3 (obsolete!)

MC Benchmark V0.1 (obsolete!)

The results of the v0.1 and v0.3 benchmark give a good overview and are retained for historical reasons. Since the testing was different, results cannot be compared directly.