MacBook M1 Processor - Audio Software Performance

November 30, 2020

Share


After the release of Apple's new M1 processor along with the new MacOS version Big Sur we received a lot of questions regarding compatibility with our software. In addition, the new processor which features its own "Neural Engine" module is obviously very interesting to us and sounds like promising news to further explore our neural network based audio processing approaches. In this article we want to share the results of the first compatibility tests and also the performance of the M1 when processing machine learning based audio algorithms.

Speed Comparison of Apple M1 versus Intel i9



Apple's ARM based M1 Processor

Currently, there is a huge hype around the topic of Apple moving away from Intel chips and introducing a proprietary ARM-based chip. The first feedback of people testing the new devices is pretty positive and especially the increased processing speed while having a lower power consumption at the same time sounds like a very pomising development. Obviously, we are interested in how our audio plugins perform on the new architecture and want to make sure that there are no compatibility issues.

Our test machine is the base model of the MacBook Air - in other words the M1-machine with the lowest specs available. In a first test, we ran our noise-reduction plugin VoiceGate using the currently available installer without any modifications. Thanks to the Rosetta translation layer, it is possible to run programs which are compiled for Intel processors also on the M1 processor which worked without any issues for our plugins. In the next step we recompiled VoiceGate to be optimized for the new processor. The following chart shows a processing speed comparison of the MacBook Air M1 with an MacBookPro 16inch Intel i9:

Speed Comparison of Apple M1 versus Intel i9

Here, we compared the time VoiceGate needed to process audio buffers of size 1024 samples. Suprisingly, even with the non-optimized version and the translation layer in between, the MacBook Air M1 is faster than the Intel i9. The optimization futher increases the performance so that VoiceGate runs almost twice as fast on the MacBook Air compared to the higher spec MacBook Pro 16inch. This was rather unexpected and very impressive. However, in practical use-cases this difference probably won't be that noticeable since the processing of both machines is pretty fast and below 1ms per buffer. Only on very high loads with lots of plugin instances in parallel the speed difference would probably matter.

As of now, we didn't explicitly use the "Neural Engine" module which is integrated in the processor. However, for future development this might be very helpful to increase the capacity and capabilities of our neural network based audio processing algorithms. Currently, the computational requirements are still one of the main limiting factors in the development process.


Big Sur

In contrast to the last MacOS release Catalina, our software luckily didn't show any compatibility issues. The accentize installers which are currently available worked without any problems even on the M1 machine. Nevertheless, if you experience any undesired behaviour of our plugins with Big Sur please let us know at support[at]accentize.com.

The M1 optimized versions of all our software will be available soon.