It would also be useful if there were Mathematica-specific features included in the Benchmark.

Right now the internal benchmarks entirely target general and universal sorts of operations that can be compared to other softwares - things like matrix multiplication, inversion, and Eigenvalues. But I am a committed Mathematica-lover and haven't even looked at another software package since our marriage back in v.4. And I don't do all that much with matrices either, come to think of it. Hence I am mostly interested in how my old ball-and-chain is improving in the realms of ReplaceAll, Cases, Select, Table, Compile, Module, and general pattern matching. But none of these feature in WolframMark/Benchmark.

B

For further comparison:

System: Mac OS X x86 (64-bit)
Date:   September 28, 2017
Wolfram Language Version:   11.2.0
Benchmark Result:   2.98

(When originally posting this, I forgot to start with a fresh kernel. Doing so bumped the benchmark result from 2.40 to 2.98.)

Just to bump up the topic, in a cordial way..., and waiting for eventual proactivity on the centralized and versioning system, here's a benchmark on an AMD Ryzen 7 1700 eight core 3.00 GHz, with Mathematica 11.2.0 (information currently in demand by some curious minds).

Attachments:

If there are enough interested people, it would be more useful to start a community benchmarking project, with open discussion/feedback to ensure quality and versioning to ensure reproducibility.

Those example measurements in the Benchmarking package are useless. They haven't changed in ages. They are exactly the same even in the oldest version I have installed, which is 10.0 from 2014.

If you open up the package source, or evaluate

Benchmarking`Private`$BenchmarkLibrary

you will see that there are some results from 2010 / M8.0 and some form 2014 / M10.0

Mathematica's performance has improved a lot between 8.0 and 10.0, and again a lot since

Furthermore, did the benchmarks themselves change? We do not know. Thus comparisons are quite meaningless.

The benchmarking package is interesting if you compare two computers with the same M version, or compare successive versions of M (and manually verify that the benchmarks have not changed). But even then, it is probably in need of some updates ...