Some Notes and Benchmarks on the Apple M1 Mac Mini

I’ve just received my Mac Mini with an M1 processor, 8 GB RAM, 256 GB SSD. These are some notes made while experimenting with it.

Preparing

Make sure you run Software Update to get the most recent version, currently 11.0.1. If you don’t do this before installing the developer tools then you will have to repeat that.

xcode-select --install

Ubuntu for arm64 is tested using Apple’s built in virtualization framewoork, with a simple UI courtesy of @KhaosT

Example ssh session

Set-ups Being Compared

Tests reported here have been run on the following configurations. For relative performance I use the original Zen 1 ThreadRipper as my base, because that’s what I sit at all day every day and other than the number of cores it can represent the two or three year old machine that most people probably use.

Name System Cores RAM GB Host OS Environment

ripper

ThreadRipper 2990wx

32 + HT

128

Ubuntu 18.04

x86_64 native

thinkpad

ThinkPad E14 Gen 2 Zen 2 4500U

6

8

Windows 10

x86_64 Ubuntu 18.04 in WSL 2

mini

Mac Mini M1

4 + 4

8

MacOS Big Sur

arm64 native

rosetta

Mac Mini M1

4 + 4

8

MacOS Big Sur

x86_64 in Rosetta 2

minivm

Mac Mini M1

4

4

MacOS Big Sur

arm64 Ubuntu 20.04 in VM

pi4

Raspberry Pi 4 1.5 GHz

4

4

Ubuntu 18.04

arm64 native

Benchmarks

Perl Summing Numbers Benchmark

This is really trivial, but you can do it in a few seconds on almost any machine. I’ve been using it since the mid-2000s when it used to take 2 or 3 seconds on Pentium IIIs and G4s.

time perl -e 'for($i=0;$i<=1000000;++$i){$t+=$i}print"$t\n"'
System Time s Speed

ripper

0.074

100%

thinkpad

0.051

145%

mini

0.045

164%

rosetta

0.057

130%

minivm

0.041

180%

pi4

0.183

40%

Note: Ubuntu in a VM being faster than native is real because Linux is more efficient launching very short-running programs than MacOS. This is also very noticable when running ./configure for open-source software.

Primes Benchmark

I have a simple prime number sieve program I wrote and have been using for a number of years on many different systems I’ve had access to. It only uses 8k of memory so fits into L1 cache on any recent CPU.

Source code with many test results: http://hoult.org/primes.txt

System Time s Speed

ripper

3.525

100%

thinkpad

3.448

102%

mini

2.795

126%

rosetta

2.925

120%

minivm

2.810

125%

pi4

12.115

29%

Building the RISC-V toolchain

Building cross-compile GNU binutils, gcc, gdb, NewLib for the RISC-V instruction set

mkdir riscv_bin
git clone --recursive https://github.com/riscv/riscv-gnu-toolchain
cd riscv-gnu-toolchain
./configure --prefix=`pwd`/../riscv_bin --enable-multilib
time make -j`ncpus`
System Time Speed

ripper

10m51s

100%

thinkpad

19m57s

54%

mini

-

-

rosetta

39m05s

28%

minivm

15m13s

71%

pi4

102m28s

11%

Note: currently a native arm64 MacOS build of binutils succeeds, but building a cross-gcc is not yet supported.

Note: minivm was allocated 4 GB RAM and 4 CPU cores. The CPU could clearly do better if I could afford to allocate more RAM to the VM. Using more cores caused it to run out of RAM and thrash. Allocating more RAM to the VM made the host MacOS unhappy.

Note: pi4 has a pimorini fanshim.