Apple M1 Performance On Linux: Benchmarks Better Than Expected For Its Alpha State

[phoronix]

Phoronix: Apple M1 Performance On Linux: Benchmarks Better Than Expected For Its Alpha State

Last Friday the crew at Asahi Linux led by Hector Martin released the first alpha release for running Linux on Apple Silicon hardware. I eagerly loaded up Asahi Linux on an M1-powered Apple Mac Mini knowing the various early limitations of the Linux kernel support that is still settling. Overall the Apple M1 Linux performance ended up exceeding my expectations for the performance in its early alpha state. Here are some benchmarks.

https://www.phoronix.com/vr.php?view=31000

[mirmirmir]

linux being better on mac devices let's go....

[vladpetric]

Microarchitecture matters hugely, and that's a reason why a comparison with the RPi4 (despite having similar instruction sets) is not that meaningful.

RPi4 - toy-grade. Can issue up to 4 instructions per cycle.

Apple M1 - best processor on the planet at this time. Can issue up to 12 instructions per cycle (+ 4 vector instructions ... ).

[schmidtbag]

Linux was overall performing very well; this is way better progress than I expected. But I'm confused...
I thought the M1 Macs had performance similar to an 11th gen i7, yet, in these tests it is barely keeping up with an i5. What am I missing here? Perhaps it depends on which application is being used.

[cypx]

I thought the M1 Macs had performance similar to an 11th gen i7, yet, in these tests it is barely keeping up with an i5. What am I missing here? Perhaps it depends on which application is being used.

You probably refer to the first M1 review, most of the time it was compared to the latest Intel processor (i5-8500B or i7-8700B) used on Mac Mini which are a bit old.

[DavidSan]

Very interesting benchmarks...

However, as far as I know, macOS balances processes between efficiency cores and performance cores. It does not seem Linux is taking this into consideration (due to the early stage) and it is running at full energy consumption on performance cores.

[scottishduck]

Linux was overall performing very well; this is way better progress than I expected. But I'm confused...
I thought the M1 Macs had performance similar to an 11th gen i7, yet, in these tests it is barely keeping up with an i5. What am I missing here? Perhaps it depends on which application is being used.

As stated in the article the OS does not yet support power states or boosting, there’s also some lack of optimisation for AARCH64 when it comes to desktop oriented applications.

[schmidtbag]

As stated in the article the OS does not yet support power states or boosting, there’s also some lack of optimisation for AARCH64 when it comes to desktop oriented applications.

But what about the Mac OS results? If Mac OS is optimized and can tap into power states, why is it overall performing worse?

[ddriver]

What I am curios about is how does this test suite compare to the performance of their own native apps. I suspect they already have a ton of hardware acceleration baked into their libraries, thus available to the native applications which use them.

3rd party binaries are good for apples to apples comparison, but if that doesn't tap into the full potential of the soc accelerators, it will never reach neither performance nor power efficiency parity.

[andre30correia]

Macos is really a piece of garbage OS, a lot of benchmarks faster in Linux.. and The story about its the fast cpu is another big lie created only in geekbench that bench paid by apple