Support for new Apple M1 "System-on-Chip" Processors

mgirdwood

Member
Oct 17, 2014
9
1
Hi Yannis.
I tested the installation of 1.8.8.6 with the deletion of the startnet files as suggested on my M1 mac mini and so far everything seems to be working fine. i am not a heavy or particularly good user but i ran a a simple RGB integration with Cal/Align/integrate and the standard processes -crop, DBE, TGV Histogram and Curves with no issues, tried also using Weighted Batch but that failed ( although to be fair i have had this also on my iMac), using the older Batch processing it flew though with no problems.

Not particularly scientific testing i know but it gives me hope i can keep the M1 as PI is probably the heaviest program i use.

Mark
 
  • Like
Reactions: Jprovida1

Yannis Doukakis

Active member
May 19, 2020
43
4
Thessaloniki, Greece
/Applications/PixInsight/bin/StarNet-pxm.dylib
It worked fine. I now face a small issue. The A12Z bionic that DTK has, has 4 efficiency and 4 performance cores. However PixInsight detects the machine as having 4 cores only, so the efficiency ones are sleeping.
Is there any setting I can force PixInsight to use 8 processors? The setting in Preferences controls the maximum not the minimum.
See the benchmark I did on this machine - shows clearly 4 cores.
 

Attachments

Last edited:

pfile

PTeam Member
Nov 23, 2009
6,050
163
dunno about this - rosetta may only report those 4 cores to the emulated machine. do you have any other x86 software that might try to use all 8 threads to see if the problem is common to rosetta or is a PI-only probem?
 

Yannis Doukakis

Active member
May 19, 2020
43
4
Thessaloniki, Greece
dunno about this - rosetta may only report those 4 cores to the emulated machine. do you have any other x86 software that might try to use all 8 threads to see if the problem is common to rosetta or is a PI-only probem?
I am currently testing with my applications, also Rosetta and Qt. WIll report back to you soon.
 

ben-pix

New member
Dec 3, 2020
2
0
Hi all, PixInsight newbie, but I just wanted to report that - having removed /Applications/PixInsight/bin/StarNet-pxm.dylib - I have PixInsight working very happily on my new Apple MacBook Air (13 inch) M1 chip, with 16gb memory, i.e. the fanless model. So far this week I've run blink, debayer, star alignment, image integration, STF, crop, DBE, ABE, Background neut, CC, PCC, histogram transformation, multi scale linear transform, and I've had no issues at all. I'm only beginning and not a large stack (just 7 lights), but the MacBook's also running a 4K display, YouTube videos, all my files are on OneDrive (including pixinsight) and are on constant back-up, and I have a word doc open documenting my steps, but as I say it's all good. At start-up the process control tells me its using 8 logical processors, with thread CPU affinity control enabled. It all feels pretty 'zippy', and seems to be working at the same sort of pace as the tutorials I'm watching, if not faster. Just thought that might help. Would be great if the star-net issue might be addressed, because if it were then PI would be fully-functioning 'out of the box' on M1, at least for a newbie/amateur. (But then only those in that category would be using such a machine for this!)
Thanks to PixInsight for all the hard work,
B
PS. PI actually has the M1 to thank for my trialling PI: I was using DSS through Parallels for stacking on my last MacBook Air, but Parallels has not yet released an M1 compatible version, so I had to look elsewhere...
PPS. Happy to trial any amended version of StarNet to see if PI can then load.
PPS. I've attached the PI benchmark test, having just run it in 'real-life conditions', i.e. after 3 hours up-time, 4k display running, OneDrive, outlook and word open etc.
 

Attachments

Last edited:

Yannis Doukakis

Active member
May 19, 2020
43
4
Thessaloniki, Greece
Thanks a lot for the benchmark. It seems that the actual M1 is properly detected as 8 cores, the issue that I reported previously for A12Z can be disregarded.
M1 seems to be (at least for PixInsight) double speed than A12Z... that IS impressive. I expected something in the 20-30% range...
Does it show you the graphs in Subframe Selector properly?
 

Marcelofig

Member
Dec 30, 2017
11
1
Thanks for the report, PI in M1 looks promising even under Rosetta 2. Wishing the PI team change their priorities a bit and see a native version soon ;).
 
  • Like
Reactions: MountainAir

MountainAir

New member
May 16, 2020
4
1
PPS. I've attached the PI benchmark test, having just run it in 'real-life conditions', i.e. after 3 hours up-time, 4k display running, OneDrive, outlook and word open etc.
Wow. That $999 MacBook Air M1 benchmarks at roughly twice the speed of my fully-upgraded top of the line iMac from a 2016-ish... and yours is running in an emulator! I'll benchmark my top-of-the-line 2020 MBP later to see how it compares, later.

I didn't take the ThreadRipper comment above as unrealistic if you're a real pro, but those machines do run $10k nicely-equipped (over $8k to build it myself). And they run so hot that I'd need to run the air conditioning in January to keep the room livable! Instead, I think I'll wait for a 2021 M1 iMac for ~$3k (all a total guess) and still end up with 3x my current performance (or better, if PI goes native). :cool:
 
Last edited:

MountainAir

New member
May 16, 2020
4
1
<gulp>

That MacBook Air M1 running PI in Rosetta 2 emulation is faster than my MBP with an 8-core 2.4 gHz Intel Core-i9 in everything but the CPU score (which the MBP was only slightly faster in). The duration was shorter on the M1, swap performance was better, etc. Wow.
 

astrogruff

New member
May 11, 2020
2
0
It's good to know the the M1 works with Rosetta 2. Even though I have a Linux server that I do most of my processing I still have PI on my current MBP but with that finally starting to give up the ghost I will be upgrading to a new M1 based laptop (Santa just told me that the astrophotography gear he ordered for me isn't coming in so I'm getting a new MBP instead).

I would note that Rosetta 2 is not an emulator like Rosetta 1 was. For x86_64 binaries, it translate the code on install to arm64 (or over the first run of the application) and it does perform just-in-time translation for things like JavaScript compilers. Emulation is not its thing, translation is.
 

retret66

Member
Nov 13, 2018
7
1
www.flickr.com
My macbook pro M1 16GB version is coming on Jan 15, I currently have the late 2013 trash can Mac Pro but I upgraded it to the fastest NVME that it can handle and added RAM to 64GB total. I will be testing its speed stacking and aligning for 1000+ 16MP subs and see if it is faster than my current setup. I am eager to find out. Final Cut pro (latest version) however on my current setup struggles so much it is painful to watch, this alone is enough reason for me to upgrade to M1. I will then make this macbook M1 as a main desktop. I will update here with the result.
 
  • Like
Reactions: pfile

Marcelofig

Member
Dec 30, 2017
11
1
My macbook pro M1 16GB version is coming on Jan 15, I currently have the late 2013 trash can Mac Pro but I upgraded it to the fastest NVME that it can handle and added RAM to 64GB total. I will be testing its speed stacking and aligning for 1000+ 16MP subs and see if it is faster than my current setup. I am eager to find out. Final Cut pro (latest version) however on my current setup struggles so much it is painful to watch, this alone is enough reason for me to upgrade to M1. I will then make this macbook M1 as a main desktop. I will update here with the result.
We'll be on the watch. I'm thinking of buying a Mac Mini M1 and I'm very interested in seeing real results from the stacking of many subs.
 

retret66

Member
Nov 13, 2018
7
1
www.flickr.com
Here is the result:
Mac Pro Late 2013 12-Core / 1TB OWC SSD/64G RAM/D700 vs Macbook Pro M1 16GB version

Subs:
439 calibrated frames @ 65MB each with drizzle enabled.

Pixsinsight Settings:
Star align: all default, just pick the same reference, cache disabled and only use the same OS drive.
Integration: All default except for Windsorized Sigma Clipping and scale + Zero Offset

Result is somewhat surprising, given the M1 is emulating and only has 8-core vs 12-core. Star align has a big gap and I can hear the M1 fan screaming. Stacking on the other hand is pretty close. As far as editing, it is pretty snappy and I believe M1 has a little advantage. By the way I use a 6K monitor for the test.
 

Attachments

Last edited:

Marcelofig

Member
Dec 30, 2017
11
1
Thank you for posting the results. Very impressive, considering that it is under emulation and with 16GB of RAM.
 

AirBourn

Member
Nov 6, 2020
11
0
I have a new M1 MacMini and can confirm the 1.8.8-7 crash on startup with Rosetta. I can also confirm that removing the /bin/StarNet-pxm.dylib allow the app to launch.