Version for MacOS with Apple M1 processor?

Moderators: Site Moderators, FAHC Science Team

Re: Version for MacOS with Apple M1 processor?

Postby gunnarre » Sat Nov 21, 2020 10:07 am

May I propose an advanced client option to disable and/or force on AVX_2, AVX and SSE processing? Such an option could be used to test if the M1 could emulate fast enough to be useful, and it might also be a good debugging option for those who are wondering if there's something wrong with the AVX processing.
Image
gunnarre
 
Posts: 207
Joined: Sun May 24, 2020 8:23 pm
Location: Norway

Re: Version for MacOS with Apple M1 processor?

Postby tonytong » Sat Nov 21, 2020 9:23 pm

I have a new M1 MacBook Air and the latest client 7.6.21 wouldn't launch but I tried using 7.6.13 as others have suggested and it has been running fine for about 10 minutes so far. PPD started at around 45000 but is now around 22000, likely due to the passive only cooling for the Air. A MacBook Pro or mac Mini will likely have better sustained PPD.
tonytong
 
Posts: 1
Joined: Thu Mar 26, 2020 11:04 pm

Re: Version for MacOS with Apple M1 processor?

Postby Foxbat » Wed Nov 25, 2020 3:29 pm

Since the M1 has four high-speed Cores and 4 low-power cores, how would that impact inter-process communications between the CPU Core threads? Would that cause time-outs in processing, or does Rosetta 2 hide those details from the program being run?
Image
Foxbat
 
Posts: 94
Joined: Wed Dec 05, 2007 11:23 pm
Location: Michiana, USA

Re: Version for MacOS with Apple M1 processor?

Postby CharonPDX » Thu Nov 26, 2020 10:57 pm

Well, running the Intel F@H on my new M1 Mac Mini, it runs through Rosetta 2 with 8 threads, and runs faster (in Estimated PPD) than the Core i9-9880H in my 2019 MacBook Pro using 16 threads.

Over the last week, I've been averaging 60,000-70,000 PPD on the M1. My Core i9 is averaging 30,000-50,000.

Because of the way Rosetta 2 "translates" code ahead of time to native code then runs the native code, rather than conventional run-time emulation, that should be pretty close to native performance. (Other benchmarks show a ~5% performance hit using Rosetta 2.)

It would be nice to have F@H aware of high-speed vs. low-power, and only run on the high-speed. And, of course, running on the GPU would be great, but that would be rather significantly more work.
CharonPDX
 
Posts: 1
Joined: Thu Nov 26, 2020 10:42 pm

Re: Version for MacOS with Apple M1 processor?

Postby MeeLee » Sun Nov 29, 2020 2:55 am

Almost the same issue on some Boinc projects.
It seems to start the WU on the performance cores (in which the program works), but then tries to offload the WUs to the higher efficiency cores, at which it can't process the data.
On Boinc for Android, you can only select the amount of low performance cores. If your device has 2 or 4 high performance cores, they're only used to load the program, and cause projects that aren't written for Cortex A50 series cores to work, to freeze...
Could be related..
MeeLee
 
Posts: 1143
Joined: Tue Feb 19, 2019 11:16 pm

Previous

Return to V7.6.x Public Release Windows/Linux/MacOS X

Who is online

Users browsing this forum: No registered users and 1 guest

cron