1
$\begingroup$

I just got a new PC with a Intel(R) Xeon(R) E-2274G CPU @ 4.00GHz processor with 4 cores and loaded ver. 12.1.1. I notice that I receive different Benchmark results before and after I launch the parallel kernels and was wondering if someone could explain this.

Is my system going to the cloud to run the benchmark after I launch the kernels and, if so, why? What is the "4-node homogeneous" cluster mentioned in the second benchmark?

I did notice that, the first time I launched the kernels, Mathematica seemed to access several web sites, but it only did this once. Maybe this is related.

Needs["Benchmarking`"]
Benchmark[]

(* Out: 
{"MachineName" -> "dom", "System" -> "Microsoft Windows (64-bit)", 
  "BenchmarkName" -> "WolframMark", "FullVersionNumber" -> "12.1.1", 
  "Date" -> "December 15, 2020", "BenchmarkResult" -> 3.14, "TotalTime" -> 4.408, 
  "Results" -> {{"Data Fitting", 0.305}, {"Digits of Pi", 0.261},
                {"Discrete Fourier Transform", 0.35}, {"Eigenvalues of a Matrix", 0.271}, 
                {"Elementary Functions", 0.507}, {"Gamma Function", 0.282}, 
                {"Large Integer Multiplication", 0.303}, {"Matrix Arithmetic", 0.326},
                {"Matrix Multiplication", 0.224}, {"Matrix Transpose", 0.423}, 
                {"Numerical Integration", 0.425}, {"Polynomial Expansion", 0.06}, 
                {"Random Number Sort", 0.179}, {"Singular Value Decomposition", 0.22}, 
                {"Solving a Linear System", 0.272}}
} *)

LaunchKernels[];
Benchmark[]
    
(* Out: 
{"MachineName" -> "4-node homogeneous cluster", "System" -> "Windows-x86-64", 
 "BenchmarkName" -> "WolframMark", "FullVersionNumber" -> "12.1.1", 
 "Date" -> "December 15, 2020", "BenchmarkResult" -> 6.09, "TotalTime" -> 27.273}
*)
$\endgroup$
7
  • $\begingroup$ A different benchmark is run depending on the system type. In the case of parallel kernels existing, the benchmark run is one specifically for "clusters" - that is, for example, multiple kernels running on the same computer. If you remove the semicolon from LaunchKernels[], you'll see that four kernels are spun up (to work with your number of cores, I suppose). This is the 'four-node homogeneous cluster' you're seeing. $\endgroup$
    – Carl Lange
    Dec 15, 2020 at 21:30
  • 2
    $\begingroup$ A small note - it seems like the benchmarking functionality is pretty neglected (if you run BenchmarkReport, the comparable results seem like they are from circa 2012), so I wouldn't rely on it too much! $\endgroup$
    – Carl Lange
    Dec 15, 2020 at 21:31
  • $\begingroup$ Thanks for that Carl. Can I ask which one is correct? That is, is the first benchmark of 3.14 correct or the cluster at 6.1? Which one should I report in the thread of benchmarks we have here then? $\endgroup$
    – Dominic
    Dec 15, 2020 at 21:45
  • $\begingroup$ I imagine the first one. The number of kernels you can have running is limited by your license, so it would be unrepresentative of the hardware (for instance I have a 24-core machine but can only run 8 kernels). $\endgroup$
    – Carl Lange
    Dec 15, 2020 at 22:13
  • 2
    $\begingroup$ Well, I've spent some time with the benchmarking threads here and my conclusions are, with all due respect to Wolfram and the fine software they have developed, that the entire benchmarking program is quite a bit sloppy: poorly documented, ambiguous reporting, poorly formatted, and unorganized to the point of where I just say, "screw it" and not even interested in reporting my results. Really should have a more professional, rigorous presentation. Very disappointed in this particular portion of Mathematica. $\endgroup$
    – Dominic
    Dec 16, 2020 at 12:59

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.