Anybody here tried to run BOINC on a new MacBook Air, Pro or Mac Mini with the M1 CPU/GPU chipset yet? Any output rate or notes you can share? Looking at the Mac Mini M1 model as a low power option for a farm for BOINC work.

This is a three-part problem.

1) Very few people have developed or fully tested new software to get the best out of the M1 chip yet. There is supposed to be a transitional emulation package to enable the new Macs to run older software until the transition is complete, but speeds and efficiency are likely to be lower than theoretically achievable.

2) The developer of BOINC for the Apple platform, Charlie Fenton, has made sterling efforts to ensure that a version of the BOINC package suitable for the new environment will be available as soon as possible after launch. I think it's nearly there, but he wants to wait for test users to report their initial findings before making a public release.

3) Once the machines and BOINC are available, individual science projects will have to update their applications as well. This is likely to be the longest and slowest stage of all, and also the one where the temporary use of the emulation layer has the most (bad) effect.

Overall, it's a big job. Early reports will be interesting, but don't interpret them as the final answer.

Edit - if you're feeling adventurous, look at the 17 November entries at the top of https://boinc.berkeley.edu/dl/?C=M;O=D

Performance wise, the numbers Apple produced make no sense.
Their CPU is benchmarked like a 7th gen Core i7 (4c/8t) at 3Ghz, yet they continuously try to compare it to a 10th or 11th gen Intel CPU in their charts.
Their GPU is an 8 core, 128 shader GPU, running at ~2-3Ghz, yet Apple tries to compare it's performance to a GTX 1050Ti ; which makes no sense, as that GPU has only 6 cores but has 768 shaders (6x more).
A lot of their benchmarks are also peak performance, after which the chip will heat up (in passively cooled systems), and the turbo boost frequencies will drop.
Unlike Nvidia GPUs, that are rated at their factory boost frequencies (which are often 100-400Mhz lower than what the aftermarket version of the GPU does).

My own research shows that I'd be better served going with a last gen. Ryzen/Intel with last gen. GPU (NVIDIA RTX2000 series, etc) as it would be around the same cost, be upgradeable and expansion ready.

Your comparison of the Apple & nvida GPUs using core & shader counts is not really valid as not all cores and shaders are created equal. What have type of work Apple targetted their GPU at? How fast does it work in "real world" applications? Will it run something like the Blender benchmark? Why worry if they run their benchmarks at peak when everyone else does the same?

https://gizmodo.com/the-macbook-air-was-a-fine-laptop-but-apples-m1-chip-m-1845671122
You know a Mac is impressive when it dazzles a die-hard PC user. The M1's 8-core CPU, with four cores devoted to performance and four to efficiency, is truly something, even compared to competing 8-core chips. On Geekbench 5, a basic test of overall system performance, the MacBook Air’s 1712 single-core and 7441 multi-core scores easily blazed past the new Dell XPS 13, which sports Intel’s top-of-the-line Core i7-1165G7 processor and starts at $1,500 (the configuration we tested is $1,600). The Dell notched a 1214 single-core and 3833 multi-core score. We also stacked the Air with M1 against a comparable machine built on one of AMD’s best laptop chips, the Ryzen 7 4800U, which has 8 cores/16 threads and is based on 7nm chip architecture. The Lenovo IdeaPad 7 Slim’s single-core (1129) and multi-core (5478) scores were also no match for the M1.

On Cinebench R23, which is similar to Geekbench but more time-intensive and therefore a potentially more thorough gauge of CPU and GPU performance, the Air’s single core (1490) and multi-core (6931) scores again bested the Dell XPS 13 with 11th-gen Intel chip and 16GB of RAM, which notched single core and multi-core scores of 1420 and 4207, respectively. The Lenovo with its AMD chip scored 1061 (single core) and 7225 (multi-core) on Cinebench, giving that chip the edge over the M1. That’s not terribly surprising, because AMD is usually the champion when it comes to multi-core benchmarks and intensive tasks.

In Handbrake, which tests the speed of the GPU when converting a 4K video file to 1080p, the Air completed the task in 8 minutes and 52 seconds. The Dell (17:24) was no match, and even the Lenovo (9:04) lagged behind. Rendering a 3D image in Blender, the Air took 6:24 using its CPU and 7:54 with its GPU. Again, those times easily beat the XPS 13 (9:47 for CPU and 10:50 for GPU) and the IdeaPad (9:37 for CPU and 9:09 for GPU) with their competitive chips. This is particularly impressive because Blender isn’t actually optimized for the M1, which means it was running on Rosetta 2, Apple’s emulation software that provides support for Intel-based Mac apps. That meant the MacBook Air wasn’t just exceptionally faster than its competitors in Blender, but it did it while also running an emulation layer.

My own research shows that I'd be better served going with a last gen. Ryzen/Intel with last gen. GPU (NVIDIA RTX2000 series, etc) as it would be around the same cost, be upgradeable and expansion ready.

Your research need to be updated. The RTX 3000 series are very appealing.
Both an RTX 3060 Ti (rumored), and 3070 based on the GA104 core, gets nearly the same results as an RTX 2080Ti for a fraction of the cost!
The RTX 3080 is probably the sweet spot of the line. The 3090 might be too expensive for most, as it's twice as fast as a 3070, but costs like 3x more.

Your comparison of the Apple & nvida GPUs using core & shader counts is not really valid as not all cores and shaders are created equal. What have type of work Apple targetted their GPU at? How fast does it work in "real world" applications? Will it run something like the Blender benchmark? Why worry if they run their benchmarks at peak when everyone else does the same?

Because it's been mentioned, that they run the mobile versions of the benchmarks, which are different than the desktop versions.
And in some way those benchmarks are not making full use of the desktop hardware.
You really going to believe that a 30W? APU like that, is going to outperform a 100W GPU?

Your research need to be updated.

Talking about that...

an RTX 3060 Ti (rumored)

The day you posted that JayZ, Steve of GN and Linus all posted reviews of your rumored 3060 Ti compared against many other (same) cards.

I just installed boinc_7.16.14_macOSX_universal on my Macbook Pro M1 (16-512)
I did a "Run CPU Benchmarks" and got
Fri 11 Dec 01:32:57 2020 | | Number of CPUs: 8
Fri 11 Dec 01:32:57 2020 | | 4674 floating point MIPS (Whetstone) per CPU
Fri 11 Dec 01:32:57 2020 | | 29666 integer MIPS (Dhrystone) per CPU

and on my Mac mini 2014 intel dual core i5 2.6 Ghz (8-256) with 7.16.14
Fri 11 Dec 01:40:15 2020 | | Number of CPUs: 2
Fri 11 Dec 01:48:39 2020 | | 4516 floating point MIPS (Whetstone) per CPU
Fri 11 Dec 01:48:39 2020 | | 19133 integer MIPS (Dhrystone) per CPU

I did a "Run CPU Benchmarks" and got

Not sure what your point is?

I was responding to the request to post result asked by Cthulhu

I have a lot of old Mac doing Boinc,
iMac 2008, Mac Pro 2008, Mac mini 2011, Mac mini 2014, and I bought a MacPro 2010 last may (and my father has a 2019 iMac)

and I hope to switch to Apple silicon and give a well deserved retirement to some of those machine,
(and have more room in my appartement). I would like the next Mac mini or a Mac mini semi pro if such machine ever exist

I guess that we need to wait for project to be converted to apple silicon before being able to get some work from boinc
it would be nice if metal was supported because my guess is that apple with have more gpu unit in next year model.

https://www.boincstats.com/signature/-1/user/3702/sig.png

---

Just like to say that I've been running Rosetta@home on my m1 mac mini for over 3 weeks now and have not had any problems.
Boinc manager is the m1 native version but the app is the default mac 'Rosetta 4.20 x86_64-apple-darwin' so its run using rosetta 2, so Rosetta is running with Rosetta.

Credit per tasks seems to of settled to an average of 700ish which is a bit more than my old intel i5-4570R iMac, which averaged around 350.

I've been running einstein@home on my 12Z Mac mini. Using the BOINC universal build, I never get any new wu's. Using a non-universal BOINC does get me FGRPSEE wu's, which in general validate fine. Am I not getting new wu's from the BOINC universal build because I have a 12Z, and not a M1?

The Einstein project allows you to see the server's decision-making processes, in the form of the server log for the most recent contact. If you could arrange to make one more request for work with the universal build, and immediately capture the server log (click on the 'time of last contact' link for the host, on the Einstein web site), I'm sure it would help the developers work out what's going wrong.

At this stage, very early in this transition from Intel to Apple silicon, it isn't clear whether the bug lies in the new version of BOINC or at the Einstein server.

Is this what you mean?

2021-01-03 19:48:15.0940 [PID=7492] Request: [USER#xxxxx] [HOST#12848621] [IP xxx.xxx.xxx.142] client 7.16.14
2021-01-03 19:48:15.2946 [PID=7492 ] [debug] have_master:1 have_working: 1 have_db: 1
2021-01-03 19:48:15.2947 [PID=7492 ] [debug] using working prefs
2021-01-03 19:48:15.2947 [PID=7492 ] [debug] have db 1; dbmod 1482088874.000000; global mod 0.000000
2021-01-03 19:48:15.2947 [PID=7492 ] [send] effective_ncpus 4 max_jobs_on_host_cpu 999999 max_jobs_on_host 999999
2021-01-03 19:48:15.2947 [PID=7492 ] [send] effective_ngpus 0 max_jobs_on_host_gpu 999999
2021-01-03 19:48:15.2947 [PID=7492 ] [send] Not using matchmaker scheduling; Not using EDF sim
2021-01-03 19:48:15.2947 [PID=7492 ] [send] CPU: req 2419200.00 sec, 0.00 instances; est delay 671879.10
2021-01-03 19:48:15.2947 [PID=7492 ] [send] work_req_seconds: 2419200.00 secs
2021-01-03 19:48:15.2947 [PID=7492 ] [send] available disk 95.19 GB, work_buf_min 864000
2021-01-03 19:48:15.2947 [PID=7492 ] [send] active_frac 0.999969 on_frac 0.995646 DCF 0.310184
2021-01-03 19:48:15.3065 [PID=7492 ] [mixed] sending locality work first (0.5816)
2021-01-03 19:48:15.3137 [PID=7492 ] [send] send_old_work() no feasible result older than 336.0 hours
2021-01-03 19:48:16.4856 [PID=7492 ] [version] Checking plan class 'Lion'
2021-01-03 19:48:16.4894 [PID=7492 ] [version] reading plan classes from file '/BOINC/projects/EinsteinAtHome/plan_class_spec.xml'
2021-01-03 19:48:16.4895 [PID=7492 ] [version] OS version required max: 199999, supplied: 200300
2021-01-03 19:48:16.4895 [PID=7492 ] [version] no app version available: APP#53 (einstein_O2MD1) PLATFORM#10 (x86_64-apple-darwin) min_version 0
2021-01-03 19:48:16.4897 [PID=7492 ] [mixed] sending non-locality work second
2021-01-03 19:48:16.5129 [PID=7492 ] [send] [HOST#12848621] will accept beta work. Scanning for beta work.
2021-01-03 19:48:16.5249 [PID=7492 ] [version] no app version available: APP#19 (einsteinbinary_BRP4) PLATFORM#10 (x86_64-apple-darwin) min_version 0
2021-01-03 19:48:16.5249 [PID=7492 ] [version] Checking plan class 'FGRPopencl-ati-mav'
2021-01-03 19:48:16.5250 [PID=7492 ] [version] parsed project prefs setting 'gpu_util_fgrp': 0.500000
2021-01-03 19:48:16.5250 [PID=7492 ] [version] No ATI devices found
2021-01-03 19:48:16.5250 [PID=7492 ] [version] Checking plan class 'FGRPopencl-nvidia-mav'
2021-01-03 19:48:16.5250 [PID=7492 ] [version] parsed project prefs setting 'gpu_util_fgrp': 0.500000
2021-01-03 19:48:16.5250 [PID=7492 ] [version] No CUDA devices found
2021-01-03 19:48:16.5250 [PID=7492 ] [version] no app version available: APP#40 (hsgamma_FGRPB1G) PLATFORM#10 (x86_64-apple-darwin) min_version 0
2021-01-03 19:48:16.5251 [PID=7492 ] [version] Checking plan class 'FGRPSSE'
2021-01-03 19:48:16.5252 [PID=7492 ] [version] CPU [' '] lacks feature ' sse '
2021-01-03 19:48:16.5252 [PID=7492 ] [version] no app version available: APP#46 (hsgamma_FGRP5) PLATFORM#10 (x86_64-apple-darwin) min_version 0
2021-01-03 19:48:16.5328 [PID=7492 ] [debug] [HOST#12848621] MSG(high) No work sent
2021-01-03 19:48:16.5329 [PID=7492 ] [debug] [HOST#12848621] MSG(high) see scheduler log messages on https://einsteinathome.org/host/12848621/log
2021-01-03 19:48:16.5329 [PID=7492 ] [debug] [HOST#12848621] MSG(high) Binary Radio Pulsar Search (Arecibo) is not available for your type of computer.
2021-01-03 19:48:16.5329 [PID=7492 ] Sending reply to [HOST#12848621]: 0 results, delay req 60.00
2021-01-03 19:48:16.5330 [PID=7492 ] Scheduler ran 1.443 seconds

2021-01-03 19:48:16.5251 [PID=7492 ] [version] Checking plan class 'FGRPSSE'
2021-01-03 19:48:16.5252 [PID=7492 ] [version] CPU [' '] lacks feature ' sse '

Correct. Your cpu is not detected as having the requisite SSE cpu capability.

Yup, that's the sort of gotcha that happens when a manufacturer makes a big design change and everybody else has to scrabble to catch up.

Streaming SIMD Extensions (SSE) is a single instruction, multiple data (SIMD) instruction set extension to the x86 architecture, designed by Intel and introduced in 1999 in their Pentium III series of Central processing units (CPUs)

And in the twenty years since then, SSE (and many subsequent evolutions later, through AVX) have become the bedrock of high performance computing. Now it's back to the drawing board.

Edit: reported upstream as #4137. Many thanks for the report.

Interesting answers to that question on an Apple forum: https://apple.stackexchange.com/a/359647
https://apple.stackexchange.com/a/407282

You (Richard) may want to add those to your ticket. I don't have my Github login on my phone.

Are there any projects that have actually produced native apps for the M1?

I have one now and would be happy to help develop either BOINC or apps.