At WWDC 2020, Apple announced it would begin transitioning its Macs from Intel to its own custom chipsets, which it referred to as “Apple Silicon”. At the time, Apple said the first Apple Silicon Macs would be released in late 2020.
Hence my question: What is Wolfram’s timetable for porting Mathematica to Apple Silicon?
Since a lot of Wolfram employees use macs, and that they closely collaborate with Apple, I’m quite confident that they will release it as soon such a mac would be available. They did the same when Apple transitioned towards Intel, they even got stage-time during the developer keynote back then…
It would be better to have some form of ‘official’ announcement. I am hoping that it may come at the WTC that starts tomorrow. If not, there are ‘office hours’ where the question will almost certainly come up. Tied in with basic functionality is whether Wolfram intends to take full advantage of the hardware – specifically using the GPU for neural nets, etc. instead of relying on NVIDIA,
Yes, I was hoping for some official info. from Wolfram, which is why I posted here. Will you be participating in those office hours? If so, please let me know if you hear anything.
It would be interesting to hear about possible GPU computing on Apple’s new hardware. As you may know, Apple stopped using NVIDIA GPUs in 2016.
I already signed up to several of them. I ‘attended’ one on Thursday last week (just to make sure I had all the extra apps I needed), and it was a zoom session. So far, there are none with Stephen, but I will keep a look out.
I spoke with someone in Tech support at the WTC. Wolfram will fully support Mathematica under Apple Silicon when it is finally released.
Stephen, in his keynote, said that for Bug Sur, only Mathematica 12.1 (and 12.1.1) and 12.2 will be supported. So, anyone using 12.0 or earlier will have to upgrade.
I still use 12.0, because 12.1 breaks my workflow. I will abstain from Big Sur too, then.
Not sure why 12.0 is not supported, because the major change for macOS – the 64 bit (cocoa) front end, is in 12.0. It is possible that Stephen missed it (or I did – I’ll check the video). Things may still be fluid, so a message to tech support may help.
I am running 12.2 beta on my Big Sur beta, and it’s pretty impressive. Lots of new stuff and fixed up stuff.
Even if you want to wait, the 12.1.1 update had hundreds of bug fixes, some of which may be of interest to you.
My problem with 12.1 and 12.1.1 is that it breaks the keyboard shortcuts that I use all the time while writing code. It also takes away the functionality of vector pdf export of 3D graphics, even wireframe ones.
The new additions are welcome, and I do keep 12.1.1 around for the occasion I may need them. But typing into 12.1.1 is an irritation for me.
Thanks for that info. George!
I forgot to ask earlier, but do you know if it will be running natively, or through the Rosetta2 auto-compiler?
How will WRI solve the problem with MKL library replacement? I am afraid that there is no comparable replacement so far. The Intel MKL library is really highly optimized for x86 CPUs, so I expect a significant decrease of numerical (float) computing performance on Apple Silicon (ARM) CPUs.
Comparable NVIDIA CUDA replacement for recent and future Apple HW is nearly impossible, because AMD (or any other) GPUs does not definitely support numerical computing on the same level as NVIDIA CUDA.
I think WRI will have to resign, at least temporarily, to the effective use of GPU on Apple HW.
This will certainly be tricky… not sure how this is now done in the countless games that run on iPhones and iPads and so on…
Actually–there is Mathematica kernel running on the iPad already, so I guess they already figured it out somehow.
One thing is “kernel running on iPad” and other thing is “kernel running effectively on iPad”. iPad numerical performance is a bit crippled.
See this report, for example.
But, to be fair … All performance issues with ARM (Apple Silicon) CPUs can of course be solved. The only question is: when?
All my Mathematica keyboard shortcuts are implemented through Keyboard Maestro, and upgrading to 12.1.1 has had no effect on any of them. Most of these shortcuts paste either functions or code blocks, but one of them quits the kernel by stringing together menu commands (Evaluation->Quit Kernel->Local->Quit).
Update: According to Mathematica 15.0 System Requirements and Platform Availability,
“Compatibility will be confirmed after full testing on the new Apple Silicon hardware finishes.”
So the question is now whether this will run through Rosetta 2, or natively on Apple silicon; and, if the former, is a native version in development, and what is its timetable?
Mathworks, for instance, says “An update for MATLAB R2020b to be released in December 2020 will provide support for running MATLAB in Rosetta 2.” They add “A version of MATLAB which runs natively on Apple Silicon is in development.” (https://www.mathworks.com/matlabcentral/answers/641925-is-matlab-supported-on-apple-silicon-macs?s_tid=srchtitle)
According to Mathematica Quick Revision History, native support for Apple Silicon was introduced in June 2021 with v. 12.3.1:
Any benchmark comparison 12.3 vs 12.3.1 on M1 ???
So far presented results (see here) shows surprisingly small performance improvement of the native M1 MMA version.
