MC-V0.6-Benchmarks
Benchmark Results (v0.6)
(These results are valid for program versions 740+)
By default, the table is sorted by descending parallel performance. You can expand the table and play with other sorting ideas (read only).
The columns are:
CPU: Name of CPU
Model: Model of CPU
Base Ghz: Base CPU clock frequency in GHz
Boost Ghz: Max CPU clock frequency in GHz
Cores: Number of real CPU cores.
VCores: Number of virtual CPU cores (hyperthreading).
Serial(Hz): The program is only allowed to use a singe CPU core. This gives a good baseline to calculate the parallelization advantage. A high score is an indication of clock frequency as well as the general sophistication of the CPU architecture. The spectra cache is cleared first.
Parallel(Hz): The program is allowed to use all available CPU resources. The speedup over the serial result should reflect the number of cores plus an extra bonus if hyperthreading is available. The spectra cache is cleared first.
Speedup: The ratio parallel/serial. This is the parallelization advantage and should scale roughly with the number of available full CPU cores. (Typically Slightly lower for non-hyperthreaded CPUs and slightly higher for hyperthreaded CPUs). A surprisingly small value could indicate poor system architecture (For example going from one to two cores on an AMD A6-4400M only gains 33% because the two cores share a single floating point unit). The value is calculated from the actual results, unaffected by the rounding to integer results shown in the table. NEW: A low number can also mean that the single core performance is exceptionally high. The latest generation of processors can go way above the base frequency if only one core is used as long as the thermal envelope is not reached (example: I7-8650U in table below).
TDP: Thermal design power in Watts. (Power use under maximum load).
User: initials of tester.
S/GHz: Serial single core performance normalized to the boost clock frequency. A better CPU architecture will give a higher score. This assumes that the serialized speed runs at max boost. Once all cores run in parallel, the clock might be reduced for thermal reasons.
P/TDP: Parallel Spectra/seconds normalized to TDP. Higher is more efficient.
NOTE: The benchmark utility also reports the cached performance, which is however not shown in the table. In this test, all required spectra are in the cache so only the overhead (cache lookup, retrieval, age update, transforms, and assembly of sub-spectra etc.) is measured. Since this is typically well over 1000x faster than calculating a spectrum, it is irrelevant for typical use and is infinitely fast for all practical purpose.
All times are in units of Hz (spectra/second). A spectrum is defined as a single MOMD orientation. With the current algorithm, all MOMD orientations of several spectra can be calculated in parallel. Spectrum cache for benchmarking is fixed at 4096 entries, independent of user settings.
The code allows up to 256 parallel instances. More details on the algorithms can be found here.
footnotes (marked in user column):
(a) hyperthreading was disabled in the BIOS, limiting the peak performance somewhat.
(b) overclocked
(e) Stock tuning
(f) Enabled MSI creator genie and XMP profile in bios (automatic overclocking)
(g) Some 12th+ generation Intel processors have a mix of "Performance-cores" and "Efficient-cores". The E-cores are not hyperthreaded and run at a lower clock.
(h) one core was busy doing an FPGA compile task.
