Apple M2 On Linux Performance Against AMD Zen 4 Mobile SoCs

[brunosalezze]

I also have an extremely strong suspicion this is the case, especially considering that Apple has bespoke extensions for things like AI

Performance is performance, a potential for optimizations always go both ways. These libraries will never be optimized for arm, maybe new ones, but the ecosystem maturity is a valid performance metric as well. Benchmarking all M2 tests on MacOS would be my only sugestion and measuring power draw with power meters are always more reliable than software.

[gnattu]

Such things may influence significantly some benchmarks for scientific computing, multimedia applications and data compression or encryption, but they do not have any influence on other benchmarks, e.g. the code compilation benchmark (won by AMD in this case).

This test for M2 is also slower than it should be. I ran the build test in a VM with only 4 vCores, and I'm getting 108s with an M1.
And Michel gets 114s with 8 cores(4p4e) and faster cores and without virtualization overheads.
Probably due to the cpu frequency driver the faster cores actually underperforms and makes them slower than a VM on M1?
I follow the test suit to create build.sh like the following and timed it:
```
#!/bin/sh
mkdir build
cd build
../configure
make -s -j 4 2>&1
```
This is the compiler in the VM:
```
gcc --version
gcc (GCC) 11.3.1 20221121 (Red Hat 11.3.1-4)
Copyright (C) 2021 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
```

And this is what I get:
```
real 1m48.146s
user 5m53.237s
sys 0m39.618s
```

[kylew77]

I'm sorry Michael has such a hard time funding the site. I for one have bought premium every time he runs a sale on it since I graduated college and got a job and before I visited the site without an ad blocker despite the ads being quite intrusive at times. I truly hope this site never shuts down. It is the best open source news site out there!

[andyprough]

Imagine that - the M2 sucks compared to real chips. Almost like it's an overpriced cell phone chip.

[oleid]

Regarding the adblock thing, I have paid for lifetime premium but tend to browse without logging in (with ublock origin in use). Wonder how many others are doing the same.

Me too.

[drakonas777]

I see a lot of coping mechanisms coming from the M lovers...

First of all, the support for accelerators is irrelevant for comparing general purpose cores. Second of all, you should back up the the "lack of optimizations" claims by the facts per specific software basis, because otherwise it's sounds just as a general excuse for an unimpressive performance which for some reason is expected by M lovers. And third of all, it's precisely as close Apples vs Apples comparison as it can be for comparing general purpose cores, at least for non multimedia-specific software, which basically runs on the generic ISA subset.

And no, there is no fucking way M2 uses 5 watts on the full all core load (adequate cooling provided), so efficiency won't be that impressive in this particular use case as well. M has efficiency advantage mostly in a sleep states, idle state and light workloads.

[partcyborg]

My issue is that I browse the site with javascript disabled and only ever have to enable it if I need to make a comment (which is something I quite rarely do), and the ads simply don't show up with javascript disabled.

I have a similar issue. I could disable uBlock Origin but still won't get ads because I also have a root based adblocker on my phone (routes ad domains to localhost via /etc/hosts). Disabling that is kind of a pain because it is device wide and I forget to re-enable it

[Steffo]

Regarding these benchmarks, everyone knows i consider Asahi a complete waste of time, both for the developers and any potential users.

If you want to have an after-life-support after Apple doesn't support your hardware anymore, then you will value this project.
As an M2-owner, I'm happy this project exists and I'm curious how well my hardware will be supported from Linux, after Apple ditches the support.

[mdedetrich]

Performance is performance, a potential for optimizations always go both ways. These libraries will never be optimized for arm, maybe new ones, but the ecosystem maturity is a valid performance metric as well. Benchmarking all M2 tests on MacOS would be my only sugestion and measuring power draw with power meters are always more reliable than software.

This is just wrong attitude, at least if you are comparing apples to apples. A lot of Linux code has had decades of optimization for specific x86/x64 instruction sets, this isn't the case for ARM especially when it comes to bespoke instruction sets which ARM supports.

Actual MacOS/software that is specifically compiled for Mac uses all of these instructions sets have insane performance.

The fact that M chips managed to so well even though a lot of the time it was using the extensions actually shows how impressive it is.

[drhoho]

Had to have Michael send me a password reset so I could login. Never would get emails to reset password...

So I was using an ad blocker.