Apple M1 SoC

[M344587487]

The M1 pro and M1 max have been announced, with ~200GB/s and 400GB/s memory bandwidth respectively (dual/quad channel LPDDR5) and up to 8 high power cores with 2 low power cores. They'll be used in laptops at least. They upped the bandwidth to feed the iGPU which has been massively expanded, the unified memory means the CPU also has access to that bandwidth which is nice.



https://en.wikipedia.org/wiki/Apple_silicon#M_series


A laptop with the 10 core pro appears to start at $2500, the 10 core max $3500. Any takers?

[rogue]

Originally Posted by M344587487

The M1 pro and M1 max have been announced, with ~200GB/s and 400GB/s memory bandwidth respectively (dual/quad channel LPDDR5) and up to 8 high power cores with 2 low power cores. They'll be used in laptops at least. They upped the bandwidth to feed the iGPU which has been massively expanded, the unified memory means the CPU also has access to that bandwidth which is nice.

https://en.wikipedia.org/wiki/Apple_silicon#M_series

A laptop with the 10 core pro appears to start at $2500, the 10 core max $3500. Any takers?

Looks like a beast. I'll be more interested in an iMac on the Mx, when it is released.

[Xyzzy]

Originally Posted by M344587487

Any takers?

Will it run (bare metal) Linux?

[M344587487]

Originally Posted by Xyzzy

Will it run (bare metal) Linux?

Work is ongoing to add support for M1 to the kernel and there's a few WIP OS's (the furthest along is probably Asahi Linux: https://asahilinux.org/blog/ ). Currently the iGPU support isn't there but it looks like Asahi can run headless or software rendered. The hope is that pro/max is similar enough to M1 that most M1 progress applies to all three. A few months after release we'll see how things stand, hopefully things go smooth and software rendering will work but I wouldn't hold your breath for the iGPU or other accelerators to be viable anytime soon.

[M344587487]

https://www.anandtech.com/print/1702...ormance-review


Apparently the bandwidth accessible by the CPU on the max tops out at ~243GB/s using all 10 cores, that's still over half the bandwidth available so at least the unified name holds true and they didn't do something odd like bolt on two extra channels only accessible by the GPU. Given that, it's hard to predict how much CPU bandwidth is available on the pro (most likely it can either fully saturate the ~200GB/s available or is ~120GB/s if half of the fabric connecting CPU to memory is trashed along with half of the memory channels).

[ewmayer]

Originally Posted by M344587487

https://www.anandtech.com/print/1702...ormance-review

At the heart of the SoC we find a new 10-core CPU setup, in a 8+2 configuration, with there being 8 performance Firestorm cores and 2 efficiency Icestorm cores. We had indicated in our initial coverage that it appears that Apple’s new M1 Pro and Max chips is using a similar, if not the same generation CPU IP as on the M1, rather than updating things to the newer generation cores that are being used in the A15. We seemingly can confirm this, as we’re seeing no apparent changes in the cores compared to what we’ve discovered on the M1 chips.

The CPU cores clock up to 3228MHz peak, however vary in frequency depending on how many cores are active within a cluster, clocking down to 3132 at 2, and 3036 MHz at 3 and 4 cores active. I say “per cluster”, because the 8 performance cores in the M1 Pro and M1 Max are indeed consisting of two 4-core clusters, both with their own 12MB L2 caches, and each being able to clock their CPUs independently from each other, so it’s actually possible to have four active cores in one cluster at 3036MHz and one active core in the other cluster running at 3.23GHz.

The two E-cores in the system clock at up to 2064MHz, and as opposed to the M1, there’s only two of them this time around, however, Apple still gives them their full 4MB of L2 cache, same as on the M1 and A-derivative chips.

Sounds like Mlucas users wanting to maximize total throughput will be best off running 3 instances, one 4-thread job per Perf-cluster and a 2-thread one on the Eff-cluster. On M1 we've found that the 4 P-cores and 4 E-cores can work together quite well via a single 8-thread instance, so possible some such hybrid will also be a competitive option on the Pro. The Max sounds bandwidth-starved, though still much better than e.g. Intel's manycore offerings.

For GIMPSers the biggest drawback of the M1 is the total-RAM maxes out at 16GB and is not user-expandable. One p-1 stage 2 running on the 16GB model can get decent performance - that allows ~300 stage 2 buffers at the current GIMPS wavefront - if nothing else is using major amounts of RAM, but with 10 or more cores one want proportionally more RAM. No word on that front in the above article, unless I missed something in the sections I simply skimmed over.

[M344587487]

The laptop offerings with pro have 16 or 32GB RAM and the max has 32 or 64GB. One other thing worth noting is that pro has 24MiB of L3 cache (labelled SLC for system level cache) which is doubled along with memory channels for max. If I had to put money on it I'd guess/hope that the CPU fabric has been specced to match the dual channel config which the quad channel config beats only slightly thanks to extra cache/memory-controllers to draw from.

[planetmike]

I'm not trying to hijack this thread, but I'm not sure where to ask this question. I got a MacBook Pro (M1) for Christmas, and for the first time in years (12?) I didn't just copy the Prime95 settings and everything over to the new machine.

I got a few dialog boxes on the initial run of Prime95, and I think I entered a bad value somewhere for how much CPU to use. Looking at Activity Monitor, only 50.5% of the CPU is in use. I don't see any options in the menus to work through to reset the CPU preferences. It looks like I only have one test running at time. On My iMac (Intel i7) I have two numbers being tested at one time. How do I set the Macbook Pro to do multiple test at once?

Apple M1, 8GB Ram. (Screaming fast machine!) Thank you for your help.

[ewmayer]

Prime95 is only optimized for the various flavors of the x86 instruction set - Apple Silicon is based on the Arm ISA. See here - note no GUI, it's command-line, and lacks some of the user features you're probably used to, like "set this to not run on battery power" - perhaps there is a fiddle in MacOS' power-saver setting which will do that for CPU-intensive tasks or selected applications.

Once you've DLed, built with "bash makemake.sh" and run the self-tests, you'll want to use of the 2 primenet.py scripts linked at the above page for work management. I wouldn't necessarily advise running a laptop this intensively, but to maximize throughput on the M1 we've found it best to just run 1 instance across the hybrid 4+4 coreset with 'nohup nice ./Mlucas -cpu 0:7 &'; that will give any p-1 stage 2 work access to at least a decent fraction of that limited installed RAM. If that runs too hot for your taste, dial it back to '-cpu 0:3' or '-cpu 0:1' and see how the temps/iteration-times behave.

PM me or post in the Mlucas SF if you have questions or encounter issues.

Originally Posted by planetmike

I'm not trying to hijack this thread, but I'm not sure where to ask this question. I got a MacBook Pro (M1) for Christmas, and for the first time in years (12?) I didn't just copy the Prime95 settings and everything over to the new machine.

I got a few dialog boxes on the initial run of Prime95, and I think I entered a bad value somewhere for how much CPU to use. Looking at Activity Monitor, only 50.5% of the CPU is in use. I don't see any options in the menus to work through to reset the CPU preferences. It looks like I only have one test running at time. On My iMac (Intel i7) I have two numbers being tested at one time. How do I set the Macbook Pro to do multiple test at once?

Apple M1, 8GB Ram. (Screaming fast machine!) Thank you for your help.