Einstein@home on M1?
21 Aug 2023 20:03:35 UTC
Topic 229981
(moderation:
Hello everyone,
Used to do Milkyway@home on my M1 macOS and it worked fine, but had to switch to Einstein, and it seems to be not really working. Got about 6000 points in almost a month. Do I have to apply some extra settings somewhere?
Thanks
Johan wrote: Hello
)
It would help immensely if your computer was un-hidden, but I will ask you this.
Do you have a discreet GPU? If so, what make and model is it?
Also, some Einstein tasks require a GPU with more than 4GB of memory, like FGRPB1G. What task(s) are you running?
Proud member of the Old Farts Association
Hi Johan, Welcome to
)
Hi Johan, Welcome to Einstein@Home!
Your computers are 'hidden' (a preference setting you could change) so nobody but you can investigate what is going on. Allowing others to 'see' your computers doesn't reveal any sensitive information.
You could spell out full details of your hardware/software plus the messages that BOINC provides when you try to get work or you could just provide a link to your host on the website so that others could see the information without you having to spell it all out. Without that information, it's impossible to know for sure what advice to give.
Cheers,
Gary.
Did not know about this,
)
Did not know about this, fixed
Johan wrote:.... had to
)
Thanks for changing your settings to allow others to see details about your hardware. I should point out that I know virtually nothing about traditional Mac hardware and even less about M1 Macs. However, I'll mention some things that seem a bit odd.
The machine shows as having 8 processors and no coprocessors (BOINC speak for GPUs). I took a look at its task list which has just 12 tasks in total - 3 completed, 4 aborted, and 5 in progress.
Both the completed and the aborted tasks show a disturbing feature - the CPU Time is only a tiny fraction of the overall Run Time. This tends to suggest that you may have the setting to suspend running when the user is active set to 'Yes' and the setting for keeping tasks in memory when suspended set to 'No'. That means that every time you use keyboard or mouse, tasks get suspended and any progress from the previous saved checkpoint gets thrown away. If you keep tasks in memory, Run Time should drop to be much closer to CPU Time.
You only need to suspend if you notice a slowdown in your normal activities. Most people don't. If you must suspend, at least keep the tasks in memory so you don't keep losing the incremental progress.
You need to go through all compute and project settings and have a good think about what each one means. If you don't understand any particular setting, you should ask for further information about it.
Cheers,
Gary.
The last couple of tasks I
)
The last couple of tasks I aborted while playing around with the settings. Checked all of those, but still get about a hundred times less calculating power than I did with Milkyway before. Seems that somehow Einstein isn't adapted to mac yet and Milkyway was
Johan wrote:... settings.
)
So what did you find?? What did you change??
You must have changed something. A previously validated FGRP5 task showed terrible values of 22898s/8892s for the two times whereas you now have a newly finished FGRP5 task that shows 7669s/7558s for the same values.
No, your computer has a certain "calculating power" that doesn't change. The thing that is different is how much more "compute intensive" the Einstein tasks actually are.
Your latest FGRP5 task took just over two hours. That is about as fast as it gets for a modern machine similar in "calculating power" to yours, irrespective of manufacturer.
Cheers,
Gary.
Unfortunately BOINC doesn't
)
Unfortunately BOINC doesn't see the GPU in the Apple Silicon chips as a GPU - it's a "generic OpenCL coprocessor" that isn't shown in the list of GPUs, and e.g. there are no preferences for this GPU in particular. However we recently published an application version for "Gamma-ray pulsar binary search #1 on GPUs (FGRPB1G)" that makes use of that coprocessor. On a M1 chip such a task should take about 32 min, which is reasonably fast.
Note that unfortunately the amount of "credit" given by each project for every minute of computing varies largely (see https://boinc.netsoft-online.com/e107_plugins/boinc/get_cpcs.php for a comparison). Don't judge the amount of computing that your machine does only by the amount of credit, in particular not across different projects.
BM
Thank you
)
Thank you