Meanwhile there are M1 Pro, M1, max, M1 ultra and M2 Apple silicon processors. A NVIDIA RTX 4090 is going to consume over 400W! So, an efficient GPU as Apple silicon GPU is, might justify an BOINC / Einstein native program development, using metal, while saving our planet! :-)
For what it's worth, PrimeGrid now can use Mac M1 and M2 GPUs for distributed computing.
I'm tried it on my M2 MBA and it seems to work well but runs warm since the MBA is fanless. Runs even better on my M1 mini and only requires the fan to run at near idle speed.
Based on the task names, it appears they are using OpenCL, not Metal.
I hoped one of the BOINC apps had converted their code to support Apple Silicon. Now that my Macs are all M1/M2-based and Apple has transitioned all its Macs to Apple Silicon, it is difficult to watch how slowly the apps run in Intel emulation mode. It's possible some Mac users aren't coming to BOINC ( or are leaving ) because of the lack of native support in BOINC's apps ( i.e. WCG, Einstein etc. ). Hopefully this support will appear. Even though Einstein doesn't have many Mac users, it might attract more if the capability were there.
I hoped one of the BOINC apps had converted their code to support Apple Silicon. Now that my Macs are all M1/M2-based and Apple has transitioned all its Macs to Apple Silicon, it is difficult to watch how slowly the apps run in Intel emulation mode. It's possible some Mac users aren't coming to BOINC ( or are leaving ) because of the lack of native support in BOINC's apps ( i.e. WCG, Einstein etc. ). Hopefully this support will appear. Even though Einstein doesn't have many Mac users, it might attract more if the capability were there.
That's always been the trade-off if you spend the money to build it will they come, with Milky Way shutting down their gpu tasks in the future that means one less Boinc Project supporting gpu's in general let alone Mac's.
For what it's worth, PrimeGrid now can use Mac M1 and M2 GPUs for distributed computing.
I'm tried it on my M2 MBA and it seems to work well but runs warm since the MBA is fanless. Runs even better on my M1 mini and only requires the fan to run at near idle speed.
Based on the task names, it appears they are using OpenCL, not Metal.
I wonder how they (PrimeGrid) do this. AFAIK the BOINC client doesn't yet detect this GPU, and thus can't pass the device specification to the app. Do they just assume that every Apple Mx chip has such a GPU and use it without management or control by the BOINC client? Or did I miss some recent development in BOINC?
I just learned that the BOINC Client does detect and support the Apple-M GPUs via OpenCL. It may require a bit of work on our server code to deal with it, though.
The OpenCL App I tested on an M1 failed with a pretty common error ("UNSUPPORTED (log once): createKernel: newComputePipelineState failed") so we probably need to change our App OpenCL code as well.
- The BOINC Client does support the Apple M GPU, but only as a "generic OpenCL coprocessor", not as a GPU with all its configuration options. E.g. the client will only run one task at a time that uses this.
- There is no support for such a coprocessor in the code of the server, it has to be added by the project individually (which is a shame). I'll do this for E@H, if at all possible in a way that is flexible enough that other projects could use it, too.
- I do have an FGRP App that works on Apple Silicon and uses the GPU. However, as the GPU lacks double precision support, part of the computation needs to be done on the CPU. Luckily enough, the FGRP App is prepared to do this.
Nope. Metal is Apple's own
)
Nope. Metal is Apple's own thing. Supported only by Apple and nobody else.
Metal IS NOT OpenCL. OpenCL is OpenCL. A open source standard supported by everybody EXCEPT Apple now.
Hi Bernd, any news here?
)
Hi Bernd, any news here?
Meanwhile there are M1 Pro, M1, max, M1 ultra and M2 Apple silicon processors. A NVIDIA RTX 4090 is going to consume over 400W! So, an efficient GPU as Apple silicon GPU is, might justify an BOINC / Einstein native program development, using metal, while saving our planet! :-)
He already stated it was
)
He already stated it was never going to happen.
They don't have the resources, financial and people to justify developing a Mac M1 gpu application for so few hosts running that hardware.
For what it's worth,
)
For what it's worth, PrimeGrid now can use Mac M1 and M2 GPUs for distributed computing.
I'm tried it on my M2 MBA and it seems to work well but runs warm since the MBA is fanless. Runs even better on my M1 mini and only requires the fan to run at near idle speed.
Based on the task names, it appears they are using OpenCL, not Metal.
https://www.primegrid.com/forum_thread. ... 207#161570
https://developer.apple.com/forums/thread/694866
https://www.reddit.com/r/BOINC/comments ... wto_guide/
I hoped one of the BOINC apps
)
I hoped one of the BOINC apps had converted their code to support Apple Silicon. Now that my Macs are all M1/M2-based and Apple has transitioned all its Macs to Apple Silicon, it is difficult to watch how slowly the apps run in Intel emulation mode. It's possible some Mac users aren't coming to BOINC ( or are leaving ) because of the lack of native support in BOINC's apps ( i.e. WCG, Einstein etc. ). Hopefully this support will appear. Even though Einstein doesn't have many Mac users, it might attract more if the capability were there.
brilor wrote: I hoped one of
)
That's always been the trade-off if you spend the money to build it will they come, with Milky Way shutting down their gpu tasks in the future that means one less Boinc Project supporting gpu's in general let alone Mac's.
BPearce wrote: For what it's
)
I wonder how they (PrimeGrid) do this. AFAIK the BOINC client doesn't yet detect this GPU, and thus can't pass the device specification to the app. Do they just assume that every Apple Mx chip has such a GPU and use it without management or control by the BOINC client? Or did I miss some recent development in BOINC?
BM
Bernd, you should ping the
)
Bernd, you should ping the PrimeGrid devs to ask how. Very interesting that Mac M1/2 can still run OpenCL.
I just learned that the BOINC
)
I just learned that the BOINC Client does detect and support the Apple-M GPUs via OpenCL. It may require a bit of work on our server code to deal with it, though.
The OpenCL App I tested on an M1 failed with a pretty common error ("UNSUPPORTED (log once): createKernel: newComputePipelineState failed") so we probably need to change our App OpenCL code as well.
BM
There's progress: - The
)
There's progress:
- The BOINC Client does support the Apple M GPU, but only as a "generic OpenCL coprocessor", not as a GPU with all its configuration options. E.g. the client will only run one task at a time that uses this.
- There is no support for such a coprocessor in the code of the server, it has to be added by the project individually (which is a shame). I'll do this for E@H, if at all possible in a way that is flexible enough that other projects could use it, too.
- I do have an FGRP App that works on Apple Silicon and uses the GPU. However, as the GPU lacks double precision support, part of the computation needs to be done on the CPU. Luckily enough, the FGRP App is prepared to do this.
BM