Will the native version of MATLAB for apple silicon macs allow me to use the GPU for calculations, as the Intel version does for NVIDIA?

The beta native version of MATLAB for apple silicon macs is now available. I will be buying a new Macbook and I am trying to work out which chip to purchase. I will be getting a chip with 10 CPU cores, but I need to decide how many GPU cores to get. If the mac GPU cores can be used for calculations, like the NVIDIA GPUs can be used with intel processors, I would pay more to have more GPU cores. Otherwise, I would buy the cheaper chip with only 16 GPUs.

Will the native version of MATLAB for apple silicon macs allow me to use the GPU for calculations, as the Intel version does for NVIDIA?

[Nur M. M.]

The GPU acceleration doesn't support on macOS is the another reason for ML enthusiasts are shifting towards opensource packages. Tensorflow supports GPU acceleration on macOS with M-series chips.

Answer 1

[Walter Roberson]

No. GPU for Apple Silicon is not being worked on, according to Mathworks. To clarify: it is not just a matter of it being in progress but not ready yet: it is not in progress at all at this time.

Side note: MATLAB for Apple Intel systems no longer supports Nvidia. Nvidia withdrew from the Apple market. Apple was refusing to approve Nvidia drivers.

The GPU use relies upon there being some robust mathematical libraries for the GPU, but those libraries are not available for Apple Silicon at this time, and Mathworks does not intend to develop such libraries itself.

[Daniel Castaño Díez]

Hi Walter,

Thanks for the info. It's really disapointing, especially as Mathworks does release beta versions of native GPU platforms for Apple Silicon:

https://es.mathworks.com/support/apple-silicon-r2022a-beta.html

Why do they offer a beta version if they don't intend to actually work in the real thing?

[Walter Roberson]

@Daniel Castaño Díez

That link makes no mention of native GPU. The beta is for the native M1-series CPU. The CPU is actively being worked on with full intent to make as many toolboxes as possible available for it -- but current plans do not include gpuArray() or GPU Coder support.

[Daniel Castaño Díez]

I see, many thanks for the clarification. Again, a real pity, and a difficult to understand decision of Mathowrks.

[Jamie Haas]

@Walter Roberson, I stumbled across this thread after learning that the 2023a pre-release for apple silicon does not support GPU. Is the position of Mathworks that native apple silicon GPU support is still not in the plan? My M1 macbook with Tensorflow is a phenominal combination of portability, battery life and impressive ML training capability. Even if there was only support for Deep Learning toolbox it would be a major improvement.

Thanks,

Jamie

[Walter Roberson]

@Jamie Haas

The last information that I had from Mathworks was that the Apple Silicon GPU will not be supported unless/until:

* there is a significant change in the marketplace notably raising the number of customers using Apple systems

* significant Deep Learning research interest in Deep Learning switches to Apple Silicon; Mathworks wants to be supporting Deep Learning researchers and developers, not simply people looking to execute code faster

* Apple significantly increases its support for GPU developers. NVIDIA is pretty accessible to people asking how to make their particular workflows faster, but reportedly Apple just gives developers the Metal interface and refuses to provide clarity on how to use the devices most efficiently

[Matthew Lord]

"* there is a significant change in the marketplace notably raising the number of customers using Apple systems"

That's unfortunate because its a self fulfilling prophecy, without the support why would anyone be motivated to use Apple systems.

Really sucks that Apple and Nvidia can't get along.

I wonder how the support will be for the new Core Ultra's since they are adding ML cores similar to Apple on their mobile chips (likely the desktop chips will also get this next).

[Jorge Ignacio Cisneros Saldana]

I do work on ML and Deep Learning research as a PhD student. I was told by apple service that Mac will be good for that mayne in next year, because IT service in my department is really bad and we cannot use university computers for doing advanced research, and was sold a M1 MAX with 64 GB back in 2023 with 10 GPU Cores. Otherwise I would have purchased a Lenovo or something else. It is 2024 and nothing yet.

[Walter Roberson]

@Jorge Ignacio Cisneros Saldana

As far as I know, Mathworks still is not working on Apple M-series GPU support for MATLAB.

[Jorge Ignacio Cisneros Saldana]

I use a lot of Matlab and Simulink for my research, I guess I will need to find alternative options at this point. Since I cannot aford buying a new windows computer for that.

[Jorge Ignacio Cisneros Saldana]

I will have to chck if i can run simualtion in simulink and then interface with python or something, I dont know at this point.