Version 3 (modified by toby, 19 months ago) (diff)


Installing a Native "Apple Silicon" GSAS-II Version on MacOS

After something like 4 months of occasional trial and much error, I was finally able to get GSAS-II to run on a ARM Mac running MacOS 11.2.3.

As I write this (6/24/21), Anaconda is not supporting ARM Macs and the wxpython version of miniforge is broken, so I had to install Python and associated packages with homebrew.

Installing Homebrew

See this page. Note that access to a Admin level account is a requirement for this. Use these steps:

  1. Install command line tools. Use command:
    xcode-select --install
  2. Log into a admin account (I used su admin in a terminal running as a standard user)
  3. Install homebrew using command:
    /bin/bash -c "$(curl -fsSL"
  1. To add brew to path, use command:
    eval "$(/opt/homebrew/bin/brew shellenv)"

Install stuff with brew

Brew is used to install a bunch of things. Here are the commands I used:

brew install svn
brew install python
brew install scipy
brew install wxpython

Install more stuff with pip

python3 -m pip install -U matplotlib
python3 -m pip install -U pyopengl  

Install GSAS-II

These steps are best done from a user account rather than an admin account. You can log out of the admin account or simply close the terminal window used before and open a new one.

eval "$(/opt/homebrew/bin/brew shellenv)"
mkdir ~/GSASII
curl -o

Compiling GSAS-II fortran routines

You probably do not need this. There is now a set of M1 (aka ARM, aka Apple Silicon) binaries in

This was fairly hard and I don't know what steps were required in retrospect. Also, it might have been better to use brew install gfortran scons rather than using miniforge to install gfortran. (I downloaded from Miniforge3-MacOSX-arm64 link on

bash ~/Downloads/
source /Users/toby/miniforge3/bin/activate
conda create -n buildg10.2 gfortran=10.2 gfortran_impl_osx-arm64=10.2 scons numpy

I installed all of Xcode to get file headers, but that may not have been needed. I had a lot of problems getting gcc to find them, but that should now be changed in Sconstruct

Speed Benchmarks

These comparisons are for two very different powder refinement cycles with a 2.3 Mh quadcore I7 MacBook Pro (early 2020) with a M1 MacBook Pro (late 2020); Both have 16 Gb.

Intel I7 Apple M1 I7:M1 ratio
4 cores 4 performance
& 4 efficiency
6.99 s 3.85 s 1.82
106.9 s 57.34 1.86