Thank you for the detailed post!!

I agree that it would be great for BOINC/WCG to update the app so it's more tuned for the M1 architecture. And this is true for any Mac app now.

Any chance you would mind running Mapping Cancer Markers for 3-4 days and let us know how many WUs it is returning per day?

Lastly, it's exciting to see how the next release (late 2021 I presume) of this chip architecture performs. I would hope to see raw performance increased 20% or more given Apple's history of the iOS chip releases. We have a mid 2017 iMac with an i7 that gets about 60 WUs a day and I was never super impressed by that number (and the fan is loud). It will be interested to see what a new iMac with an Mx chip looks like next year (and hopefully silent).

I have an initial update on the Mapping Cancer Markers testing on the base M1 Mac mini with 8GB of memory. On the M1 Mac mini, when I switched it over to exclusively running MCM1 WUs, I started to see the sporadic behavior that @ericinboston reported earlier with the base M1 MacBook Air with 8GB of memory as well - with no other apps running, sometimes it only ran 4-7 threads instead of 8. As was previously mentioned, I also observed the MCM1 WUs each use 500-520 MB of memory while running and thus for 8 threads to run simultaneously, 4,000-4,160 MB of memory needs to be available. When I was previously testing with the OpenPandemics project, the OPN1 WUs only needed 1,200-2,160 MB of memory total with all 8 threads running as those WUs only needed ~150-270MB per thread. Somewhat counter-intuitively, through opening certain additional programs selectively, I was able to coax the macOS Big Sur operating system into moving some background programs out of running memory into the Swap space and for other programs’ memory to be compressed. When I did that, the BOINC agent was then able to start running all 8 threads of MCM1 WUs successfully and for the past 12 hours or so, all 8 threads have been running at 100% consistently.

Due to initially seeing the same sporadic behavior that @ericinboston observed on the base M1 Mac mini with active cooling, I think that the data suggests that the sporadic behavior seen earlier with the M1 MacBook Air with 8GB of memory back on November 22nd was more memory utilization related and possibly not thermally related at all. While the M1 MBA would slow down the M1 chip after a few minutes of sustained load due to it not having a fan to help dissipate the heat, from the point of view of the BOINC client I don’t believe thermal throttling of the M1 chip should have made CPU cores/slots fully “disappear” from the point of view of the BOINC/WCG client - it should have just slowed down the cores themselves until the thermals returned to equilibrium.

So in summary, I think anybody with a base M1 Mac may start to bump into the 8GB memory ceiling trying to keep all 8 CPU cores running 100% on the more memory intensive World Community Grid projects. I’m guessing that these programs running through the Rosetta 2 emulation/translation layer also may use a little bit more memory per thread than on an Intel macOS machine. This system requirements page seems to list the estimated amount of memory per thread for the different projects: https://www.worldcommunitygrid.org/help/viewTopic.do?shortName=minimumreq

Africa Rainfall Project - 1 GB
FightAIDS@Home - Phase 2 - 1 GB
Microbiome Immunity Project - 650 MB
Mapping Cancer Markers - 400 MB
OpenPandemics - COVID19 - 250 MB
Smash Childhood Cancer - 250 MB
Help Stop TB - 200 MB

Now that I’ve gotten the base M1 Mac mini running the MCM1 WUs consistently on all 8 cores, I should be able to get some good data in a couple days on the average run time of those WUs with all 8 cores engaged simultaneously. Any data collected from when only some of the cores were in use would not be fully representative of average completion time since the 4 performance cores of the M1 chip are significantly more powerful than the 4 efficiency cores and thus would complete WUs much faster.

Good to know - I'm seeing that too in the results status. At the moment it looks like the majority of the MCM1 WUs I've received have been the LOO ones. I'll break those out separately.

Happy to - it was not very scientific though. I launched Google Chrome and Safari and then opened tabs until macOS Big Sur’s automated memory management features kicked in and started moving/compressing some of the running programs in memory. (That sure would be handy if there was a way to manually specify that, but I think that is fully controlled by macOS automatically.) Here is a link to a screenshot of what my memory utilization has looked like where MCM1 WUs have been running smoothly for the past couple of days:
https://imgur.com/a/2RBaD7W
Another option that you could try as well for testing purposes on your M1 MBA would be to temporarily switch that computer to exclusively doing OpenPandemic WUs (or another project with lower memory demands.) Given the memory demands of those WUs are about half that of the Mapping Cancer Markers WUs, that would help remove the memory variable entirely and allow you to observe how the MBA handles running eight OPN1 threads continuously.

I don’t have any insight into why the MCM1 WUs each need 500MB per thread on macOS. I’ve also observed those threads only needing 75-80MB on a Windows 10 machine. It is interesting that the minimum system specifications page ( https://www.worldcommunitygrid.org/help/viewTopic.do?shortName=minimumreq ) says that each thread might need about 400MB of memory. 500MB is a lot closer to 400MB than 75MB is : ) Out of curiosity, I’m going to check how much memory MCM1 threads use on an Intel Mac just to be able to compare that to an M1 Mac.

Finally, I have the results of how long the MCM1 WUs take to run on an M1 Mac mini:
The shorter LOO ‘VMethod’ type take on average 1.66 hours.
The longer NFCV ‘VMethod’ type take on average 3.83 hours.
Thus, with 192 “thread-hours” per day available (8 single-thread cores x 24 hours), it should be able to complete ~115 “LOO” MCM1 WUs/day or ~50 “NFCV” WUs/day. Right now on the M1 Mac mini I’ve been testing on, the WUs have been heavily weighted towards the “LOO” type and I’ve been seeing ~106 MCM1 WUs being completed per day.

Good memory data here. I am running an Apple Mini M1 with 8GB memory. I runs very cool, with just a slight temperature being felt.

The Africa Rainfall project is taking memory from other projects, BUT I have not selected it in my projects. It still keeps being downloaded to run.
How do i stop the Africa rainfall packets being downloaded?