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

mberetta said:
This is my benchmark with M1 Max. I think we can hope the double of the performances if UltraFusion technology works as described but Threadripper 64-core is still faster.
Click to expand...

There is also a posted benchmark for an M1 Max with 64gb memory that is about 40% faster:

https://pixinsight.com/benchmark/benchmark-report.php?sn=TH9S1KV91I9M5JBLREMSMI1DPXT0KCUQ

It will be interesting to see where things are when the M1 native version becomes available and when the new Mac Studios start showing up on the benchmark list.
 
I doubt that a native version of PixInsight will be THAT much faster, maybe 20% if we're lucky. That's still a notable gain but unless the PI team makes use of the custom processors that are in the Apple silicon the gains will probably be fairly modest. However, I think there is zero chance that they will write custom code just for the Mac (for example, I doubt that they write custom code for anyone's GPUs).

I do know that my M1 MacBook Air runs PixInsight notably faster than my Dell desktop that has a Core i7-7700 @ 3.6GHz (turbo to 4.2GHz) with 24GB RAM and multiple SSDs. But, that Dell is maybe five years old and the differences are not extreme. Geekbench suggests that the M1 should be about 1.5 times as fast as my Dell in multi-core when running native code .

The PI benchmarks for the Core i7-7700 seem to run between 8k to 9k for the CPU (under Windows), while the M1 MacBook Air is somewhere between 10k and 11k. So, if we get an 20% bump with the native version of PixInsight that might mean scores on the M1 MacBook Air in the 12k to 13k range which falls right in line with the 1.5X difference when running Geekbench native. Thus, I think the 20% improvement is probably about right. Seems like that is also what other apps are showing when they go native.

Then, if we assume another gain of 1.5X for the M1 Max and about 2X that for the Ultra we'd get a PixInsight CPU benchmark for native code that should be around 19K for the Max and perhaps in the upper 30Ks for the M1 Ultra. I did note a result that use to be in the PixInsight benchmark results that suggested the M1 Ultra was already at 28K, so again if we scale that by a 20% improvement for going native we'd get around 34K for the CPU portion of the test.
 
Last edited:
must... not... purchase... m1 ultra....mac....
Click to expand...

Why? Does this warning extend to M1 Max as well?

I had plans to get an M1 Max MacBook Pro but they cannot be found. Apple also announced the M2 in MacBooks in 2023 now? So I might wait. Or maybe just get a M1 Pro MacBook. Or maybe just keep the laptop I'm trying out right now that has an i7 11800 and RTX3060.
 
Last edited:
that was not a warning, it was me struggling against myself

i have an m1 macbook and at least 1.8.8-x worked OK there. i havent loaded 1.8.9 yet.
 
james7 said:
I doubt that a native version of PixInsight will be THAT much faster, maybe 20% if we're lucky. That's still a notable gain but unless the PI team makes use of the custom processors that are in the Apple silicon the gains will probably be fairly modest. However, I think there is zero chance that they will write custom code just for the Mac (for example, I doubt that they write custom code for anyone's GPUs).

I do know that my M1 MacBook Air runs PixInsight notably faster than my Dell desktop that has a Core i7-7700 @ 3.6GHz (turbo to 4.2GHz) with 24GB RAM and multiple SSDs. But, that Dell is maybe five years old and the differences are not extreme. Geekbench suggests that the M1 should be about 1.5 times as fast as my Dell in multi-core when running native code .

The PI benchmarks for the Core i7-7700 seem to run between 8k to 9k for the CPU (under Windows), while the M1 MacBook Air is somewhere between 10k and 11k. So, if we get an 20% bump with the native version of PixInsight that might mean scores on the M1 MacBook Air in the 12k to 13k range which falls right in line with the 1.5X difference when running Geekbench native. Thus, I think the 20% improvement is probably about right. Seems like that is also what other apps are showing when they go native.

Then, if we assume another gain of 1.5X for the M1 Max and about 2X that for the Ultra we'd get a PixInsight CPU benchmark for native code that should be around 19K for the Max and perhaps in the upper 30Ks for the M1 Ultra. I did note a result that use to be in the PixInsight benchmark results that suggested the M1 Ultra was already at 28K, so again if we scale that by a 20% improvement for going native we'd get around 34K for the CPU portion of the test.
Click to expand...
We might get more than 20%
I’ve been considering upgrading from my beloved 2013 12 core trash can and weighing the benefits of a 2x 18-core Z840 versus a 10 core Mac Studio so I went looking for comparable benchmarks. I disovered that a similar (stacking) tool’s benchmark results after going Apple Silicon native were able to deliver 100% performance improvements over Rosetta based benchmarks with an updated VM Platform.

Assuming these results are representative then it gives me confidence that I can expect similar boosts with PI and AS/Arm is looking extremely tempting compared to the Intel/AMD alternatives I’ve been considering.
 
Last edited:
Great news but this is PixInsight Forum. We are not interested in competing applications here, and comments or information about competing applications are not welcome. Please try to keep conversations strictly within PixInsight topics. See the forum rules.
 
Apologies, was trying to suggest that we might expect to see more than 20% improvements based on “similar” workloads. Re-reading my post that wasn’t clear.

move updated it I hope that helps
 
Last edited:
You must log in or register to reply here.
Back
Top