Are there any benchmarks yet for Mathematica 12.1 on the new (13") M1 MacBook Air or (13") M1 MacBook Pro?
$\begingroup$
$\endgroup$
6
-
$\begingroup$ Also, any insights on when we can expect a native ARM version of Mathematica? $\endgroup$– Chris KCommented Nov 18, 2020 at 21:32
-
$\begingroup$ Mathematica already runs on Raspberry Pi, so support for Apple Silicon shouldn't be difficult. $\endgroup$– John SmithCommented Nov 19, 2020 at 2:24
-
2$\begingroup$ @JohnSmith It seems that there isn't even a Fortran compiler yet for Apple's new platform. Most scientific software relies on Fortran one way or another. The relevant gcc/gfortran issue is "suspended" and the developers there claim that it takes several months' of work to implement support. $\endgroup$– SzabolcsCommented Nov 19, 2020 at 9:31
-
1$\begingroup$ I have had occasion to use the gcc/gfortran development branch mentioned there and it is not in such bad shape actually $\endgroup$– ilianCommented Nov 19, 2020 at 16:06
-
1$\begingroup$ There is a thread with the same question at community.wolfram.com/groups/-/m/t/2118125. There, the WolframMark benchmark was used using Mathematica 12.1.1 with the Rosetta 2 translation layer. $\endgroup$– QbyteCommented Nov 21, 2020 at 21:03
|
Show 1 more comment
1 Answer
$\begingroup$
$\endgroup$
3
Mathematica 12.3.1 as of July 2021, supports native Apple M1, my benchmark on a M1 Macbook pro, 16GB is:
{"MachineName" -> "laederlappen",
"System" -> "Mac OS X ARM (64-bit)",
"BenchmarkName" ->
"WolframMark",
"FullVersionNumber" -> "12.3.1",
"Date" -> "July 9, 2021",
"BenchmarkResult" -> 3.147,
"TotalTime" -> 4.398,
"Results" -> {{"Data Fitting", 0.191}, {"Digits of Pi", 0.171},
{"Discrete Fourier Transform", 0.307}, {"Eigenvalues of a Matrix", 0.453}, {"Elementary Functions", 0.606}, {"Gamma Function", 0.221}, {"Large Integer Multiplication", 0.187}, {"Matrix Arithmetic", 0.145}, {"Matrix Multiplication", 0.302}, {"Matrix Transpose", 0.181}, {"Numerical Integration", 0.322}, {"Polynomial Expansion", 0.048}, {"Random Number Sort", 0.399}, {"Singular Value Decomposition", 0.545},{"Solving a Linear System", 0.32}}}
Update
With a newly reset computer, and a newly reset kernel (completely fresh started mathematica) I wrote a little script to force the benchmark to be done 100 times in a row, in the hopes that the cpu would ramp up some more and give a better result:
Needs["Benchmarking`"]
results = Benchmark[] (*first run to see what it starts with cold*)
things = {};
Do[AppendTo[things, results = Benchmark[]], 100];
data = Table[
"BenchmarkResult" /. things[[i, 1, 6]], {i, 1, Length@things}];
ListLinePlot[data, ImageSize -> Large, Frame -> True, FrameStyle -> Black, GridLines -> Automatic]
Max[data]
(3.22)
Running through rosetta2, I got a max result of ~2.7...there is some improvement, but not the jump to around 3.7 or so I expected.
In this answer a user with a M1 macbook air got a result of 3.2, though they haven't said if that was their first attempt or not...I would find that somewhat strange, being that my machine has active cooling, and an extra core, while theirs has neither...
I would speculate that a different kind of benchmark needs to be developed with different work loads to really push the CPU...things mathematica used to chock on, on my other machines (even my AMD 3800x) my M1 cuts through fine...
answered Jul 9, 2021 at 21:32
-
$\begingroup$ Hm, not such a big improvement from the non-native result (2.724) you posted in another thread. $\endgroup$– Chris KCommented Jul 9, 2021 at 23:17
-
$\begingroup$ @ChrisK unfortunately not much really…but the wolfmark test is imho not great, in my minimal testing on all my machines, the used cores never use full power…today i will loop the wolfmark test 50 times and see the final result and see if the result gets any higher and report back $\endgroup$ Commented Jul 10, 2021 at 6:51
-
$\begingroup$ Maybe it's the benchmark, &/or maybe the code isn't optimized yet. That's only 7% higher than my mid-2014 4-core MBP w/ MMA 12.1.1, which scores 2.995 (I should retry w/ 12.3.1). And it's sig. slower than the 4.48 reported for a 2019 27-inch iMac with a 3.6 GHz 8-Core i9 w/ MMA 12.1.1 under macOS 10.15.7 (see community.wolfram.com/groups/-/m/t/2118125?sortMsg=Recent). This is notable, b/c the per-core speed on an M1 is generally reported as equal or faster than that on an i9—and I don't think the i9 should be advantaged by its extra cores, since I believe the benchmark uses <= 4 cores $\endgroup$– theoristCommented Jul 14, 2021 at 1:31