Apple Silicon ARM chips - M3 family support status

[kasdashdfjsah]

Does the software fully support the latest version of MacOS, as well as all the new M3 Apple Silicon chips, which have new features like ray-tracing, dynamic caching, and mesh shading? Also, the Ultra chips are composed of 2 max chips stuck together, which some software and games struggle to fully utilize, so please make sure the software supports these chips fully as well.

[Joe_H]

The v8 Public Beta fully supports Apple Silicon, both the client and CPU folding cores have been compiled o use the Apple native machine code. The rest of the features only would possibly be useful for the GPU folding. But GPU folding is not supported on any macOS system at this point.

How well the CPU folding core utilizes the Ultra processors is going to depend on the scheduler in the macOS. The same applies to any other multi chip systems from Intel or AMD running other OSs.

[toTOW]

It works, but big.LITTLE CPU architectures still require a lot of user configuration to work efficiently ...

[calxalot]

It is fastest if you do not set cpus higher than the performance core count.

First run of v8 will set this for you, unless your old config.xml has a cpus entry.

I don’t have an M3 to test, but it should work fine.

So would v7, but it will be slower because it runs in Rosetta.

[calxalot]

Some folks are using Process Lasso on Windows to deal with big.LITTLE.

No such utility is needed on macOS.
Just do not set cpus higher than the performance core count.

[Alex_Atkin]

Any chance of RustiCL support in Asahi Linux? I realise this was discussed before and the MESA drivers considered not good enough, but this may be different on Asahi which now has OpenCL 3.0 support?

[calxalot]

Does it support OpenCL 1.2 or later in C plus hardware double precision floats (FP64)?
I have seen clinfo report different version OpenCL C support.

Since Asahi is Fedora, you need to build the package yourself.
https://github.com/FoldingAtHome/fah-cl ... ING-RPM.md
The linux ARM tarball might work as-is run manually from command line.

[calxalot]

Um, I don't think there is any linux ARM GPU support anyway.

https://github.com/FoldingAtHome/fah-cl ... issues/139

[Alex_Atkin]

Never occurred to me they hadn't written cores for ARM but I guess it makes sense given how few people will use it. Kinda surprising they have an ARM port of the client at all really. I mean if they don't think its worth supporting Metal on MacOS, then what chance does Linux on ARM have.

[calxalot]

It is my understanding that Metal turned out to not provide a significant performance improvement over OpenCL in OpenMM.

If there is ever a GPU core for macOS, it will use the deprecated OpenCL.
Assuming Apple doesn't remove it.

I think the initial ARM porting was done by a third party. Neocortix?

[Joe_H]

Never occurred to me they hadn't written cores for ARM but I guess it makes sense given how few people will use it. Kinda surprising they have an ARM port of the client at all really. I mean if they don't think its worth supporting Metal on MacOS, then what chance does Linux on ARM have.

To clarify that, no GPU cores for ARM, but there are CPU folding cores that are native for ARM under Linux and Apple Silicon's ARM instruction set under macOS

[toTOW]

I think the initial ARM porting was done by a third party. Neocortix?

Indeed ... so the idea behind this port to use big ARM chips (like Ampere Altra chips with 80 cores) that you can find in datacenter/cloud platforms. Unfortunately, most ARM chips (used in SBC) are still too weak to make deadlines on FAH.

Apple M chips are the current exception to the statement above ... they might be followed by Qualcomm Elite or future Mediatek/nVidia chips soon, but we still don't know ...