Apple Silicon for scientific research

[richmlow]

Hi all,


For those who use Apple computers for scientific research, I was wondering what your feelings are about Apple Silicon for this?

I'm finding that the limited RAM in my 2023 Mac mini Pro (32GB RAM) is already being bumped against. I'm currently using the machine
for scientific computing (custom-designed Mathematica programs) in my research. I'm not really too bothered by the 10-core CPU in my machine since my Mathematica license only allows for the use of 8 cores maximum.

What are your experiences / thoughts about scientific computing and Apple Silicon? I'd be very interested in hearing!


richmlow

 

[leman]

Depends on what you do and what software you use. Mathematica in particular is not very well optimized for Apple Silicon. For some other domains (especially statistics/stochastic modeling) it’s extremely capable.

Apple Silicon brings a number of unique advantages to scientific computing. First, it can execute more scalar numerical operations simultaneously than any x86 processors, which makes it a beast on non-vectorized code. Second, it ships with very capable matrix acceleration unit (you’d need to use BLAS/LAPACK to access it).

 

[ovbacon]

Hi all,


For those who use Apple computers for scientific research, I was wondering what your feelings are about Apple Silicon for this?

I'm finding that the limited RAM in my 2023 Mac mini Pro (32GB RAM) is already being bumped against. I'm currently using the machine
for scientific computing (custom-designed Mathematica programs) in my research. I'm not really too bothered by the 10-core CPU in my machine since my Mathematica license only allows for the use of 8 cores maximum.

What are your experiences / thoughts about scientific computing and Apple Silicon? I'd be very interested in hearing!


richmlow

I really have to laugh a little about this... you are posting this as if "scientific research" is one specific thing.

You get your computer for your needs and it sounds like you didn't put in enough time researching what those needs would be.

I've done a lot of medical research on mac going back decades and have never found them to be limiting but then again my pool of "scientific research" is extremely narrow towards heart disease. If there was a one size fits all "desktop" for scientific research it would be in every scientific paper published.

It would be better if you indicate the exact needs you have or what it is you are demanding of your mac so that anyone in a similar situation can share their experience and/or setup.

 

[richmlow]

I really have to laugh a little about this... you are posting this as if "scientific research" is one specific thing.

You get your computer for your needs and it sounds like you didn't put in enough time researching what those needs would be.

I've done a lot of medical research on mac going back decades and have never found them to be limiting but then again my pool of "scientific research" is extremely narrow towards heart disease. If there was a one size fits all "desktop" for scientific research it would be in every scientific paper published.

It would be better if you indicate the exact needs you have or what it is you are demanding of your mac so that anyone in a similar situation can share their experience and/or setup.

ovbacon,


There is no need to be snarky and condescending.

I am just asking for general impressions. If you don't have anything worthwhile to add to the conversation,
please consider not opening your mouth.


richmlow

 

[Sterkenburg]

The move to AS (64 GB M1 Max) was very smooth for me. I use my Mac for science pretty much in the same way as before: local ML model prototyping, data science (mostly in Python), writing papers. The only thing I miss is proper support for local GPGPU computation but this hasn't been a viable option for a long time before the transition either so not an AS-specific woe.

 

[leman]

The only thing I miss is proper support for local GPGPU computation

Do you have some specific tools or frameworks in mind? If one only talks about custom GPGPU (as in writing your own programs), Metal is mature and easy to work with.

 

[ChrisA]

Hi all,


For those who use Apple computers for scientific research, ...
for scientific computing (custom-designed Mathematica programs) in my research.

You ask about "scientific computing" but really your question is only about Mathematica. I find that anything based on TensorFlow or Python runs well on my M2-Pro. Mathematica can be a memory hog.

Such very specialized questions might be best answered in a Mathematica forum.

Many people who need large-scale computing are doing it in the cloud. Not only can you rent time on a computer you could not possibly afford to buy, you don't tie-up your desk-top computer for days or weeks at a time while it crunches numbers.

 

[ChrisA]

If one only talks about custom GPGPU (as in writing your own programs), Metal is mature and easy to work with.

Yes, if you write your own software, the support for this is very good now. You don't need to write at such a low level. You can use tools that use the GPU. There are a few different ways to get numpy to use Apple Silicon GPU cores.

 

[richmlow]

You ask about "scientific computing" but really your question is only about Mathematica. I find that anything based on TensorFlow or Python runs well on my M2-Pro. Mathematica can be a memory hog.

Such very specialized questions might be best answered in a Mathematica forum.

Many people who need large-scale computing are doing it in the cloud. Not only can you rent time on a computer you could not possibly afford to buy, you don't tie-up your desk-top computer for days or weeks at a time while it crunches numbers.

ChrisA,


My question was not specifically about Mathematica. I just mentioned that I use it.

I'm curious about people's overall opinions/feelings about Apple Silicon and scientific computing, in general.


richmlow