Mac M1 experience with cmdstanR

General
1

Hi all,

I wonder if any M1 (or M2) user of cmdstanR can share their experiences so far. I am thinking about upgrading my Intel MacBook Pro to M1 but couldn’t quite get my arms around certain issues

  1. toolchain / setup / runtime - are any any major installation hurdles to jump through to get cmdstanr to work? Have you encountered any platform specific bugs (either related to stan or external libraries)?

  2. Memory usage - how does memory usage (both compelling and running models) compared vs x86? If I’m good with a 8Gb MacBook Pro, would I be good with 8Gb M1?

Many thanks

Andy

2

  1. No. Just be sure to download the ARM Apple silicone version of programs or the “universal” version.

  2. I advise 16 gb if you are using Stan because I advise this for any laptop if you do computations on it. I haven’t noticed any particular issue with ram being eaten more so than on a different computer. I have 32 gb and I wish it were 64 gb but depends on the amount of data and models that you intend to run.

3 Likes
3

Thanks!

4

I’ve just started using an M1 Max with 64gb, essentially an upgrade purely for running big models etc. in Stan/R! I’ve had no problems with the toolchain or setup. There are options when you install R (I think) for a specific M1 optimised version. As @spinkney suggests opt for the Apple Silicone versions of things wherever possible. I can’t compare with a previous intel Mac as I used Macs a long time ago and had been using a more powerful windows machine recently instead. On a small test I did the new laptop ran a bunch of models 2.5* faster than my previous laptop - but it could also be due to the higher RAM (previously I had 16gb, now 64)

1 Like
5

This could be of some help: Transitioning from x86 to arm64 on macOS - experiences of an R user

1 Like