Comparing the results of 4 very popular CPUs, The Apple M1 outperforms all of them, even the i9 processor in the 16” Mac

Apple M1 Neural Engine advertises 11 trillion operators per second. The question here is, is that TFLOPS (FP32)? This is compared to the Nvidia GeForce RTX 2080 Super at 11.2.

It is unclear right now if Juypter notebooks work with the Apple M1 processor. I have found one link that is running juypter notebooks on an ARM architecture using docker. Time will tell on this one. Maybe VSCode once it is production-ready will have a Juypter extension that will work.

Microsoft also announced on November 11, 2020, that it will launch a new Universal build of its Mac Office 2019 beta that supports the new Apple M1 processor. As a matter of fact, Erik Schwiebert announced on Twitter that Microsoft was releasing a beta version that same day on the beta channel. However, they did not announce any dates for the public release.

http://live-stream.munich.es/exd/v-ideos-Fenerbahce-KK-Crvena-zvezda-v-en-gb-ewr-.php

http://main.dentisalut.com/zwo/v-ideos-Skelleftea-AIK-Vaxjo-Lakers-HC-v-en-gb-1nsv-10.php

http://go.acaps.cat/npt/v-ideos-Norway-Denmark-v-en-gb-1tse-.php

http://streaming7.actiup.com/nez/video-valencia-basket-v-barca-lassa-v-es-es-1pnn-13.php

http://news24.gruposio.es/ydd/video-Fenerbahce-KK-Crvena-zvezda-v-en-gb-1vpg-.php

https://assifonte.org/media/hvc/video-brynas-v-linkoping-v-sw-sw-1jbx-15.php

http://live-stream.munich.es/exd/videos-Fenerbahce-KK-Crvena-zvezda-v-en-gb-prr-.php

http://live-stream.munich.es/exd/videos-Porto-Robur-Costa-Ravenna-GI-Group-Team-Monza-v-en-gb-rsn-.php

http://go.acaps.cat/npt/video-Norway-Denmark-v-en-gb-1jqk30122020-3.php

http://live-stream.munich.es/exd/videos-Porto-Robur-Costa-Ravenna-GI-Group-Team-Monza-v-en-gb-qiz30122020-.php

http://news24.gruposio.es/ydd/video-Fenerbahce-KK-Crvena-zvezda-v-en-gb-1bbu-5.php

http://streaming7.actiup.com/nez/video-valencia-basket-v-barca-lassa-v-es-es-1cnc-20.php

https://assifonte.org/media/hvc/videos-brynas-v-linkoping-v-sw-sw-1zzh-10.php

https://assifonte.org/media/hvc/Video-orebro-HK-Malmo-Redhawks-v-en-gb-1uyf30122020-.php

http://live-stream.munich.es/exd/videos-Porto-Robur-Costa-Ravenna-GI-Group-Team-Monza-v-en-gb-zoq30122020-.php

http://main.dentisalut.com/zwo/Video-porto-robur-costa-ravenna-v-gi-group-team-monza-v-it-it-1glb2-23.php

http://main.dentisalut.com/zwo/Video-porto-robur-costa-ravenna-v-gi-group-team-monza-v-it-it-1buu2-9.php

http://news24.gruposio.es/ydd/Video-Fenerbahce-KK-Crvena-zvezda-v-en-gb-1pzp30122020-5.php

http://streaming7.actiup.com/nez/video-valencia-basket-v-barca-lassa-v-es-es-1grf-8.php

http://go.acaps.cat/npt/video-Norway-Denmark-v-en-gb-1vul-14.php

example of what Progressier does. It also automatically tracks app installs, lets you set up rules to decide when to promote installation of your PWA, generates all the right logos, and so much more. And that’s just version 1. Push notifications, offline, caching strategies, automated app store submissions — there’s a lot of pretty cool stuff in the roadmap.

When you combined that speed with integrated memory (reduced latency) and 20 hours of battery life, that could be a game-changer for data scientists on the go. You not only get a GPU comparable to an RTX 2080 Super, a CPU that has one of the best single-core scores in the industry, a form factor of the MacBook Air, but also 18–20 hours of battery life. Wow!

Some are claiming you can perform inference from Tensorflow models by taking advantage of the Apple Converter. Here is a link to an article talking about converting from TF models to CoreML. Some others are thinking they are referring to Swift for Tensorflow.

Apple Silicon uses an ARM-based instruction set architecture. Python already runs on many other platforms including ARM, MIPS, PowerPC, i386, and x86–64. There should be no trouble with Python.

The M1 processor also brings Apple’s industry-leading Neural Engine specifically designed for machine learning processes. The neural engine has previously been added to the A-series processor on the iPad and iPhone but has yet to be on the Mac until now.

Will Scikit-learn work with the Apple M1? I think the short answer is maybe??? So, what we know is that Python will work for the ARM architecture of the M1, and Scikit-learn doesn’t take advantage of or use GPU support. So theoretically, Scikit-learn should work for the Apple M1.

Microsoft announced on November 10, 2020, on Twitter, they are working on a Universal build of VSCode for the Apple M1. As of writing this, they have the ARM experimental build upon their site located here.

The integrated GPU is an 8-core GPU, which can execute 25,000 threads concurrently. Apple is claiming this is the “world’s fastest integrated graphics in a personal computer” at 2.6 TFLOPS (FP32) throughput. For sure this is impressive for an integrated GPU but we as data scientists are interested in overall performance.

At the Apple M1 event, Apple mentioned support for Tensorflow. I have yet to find any specifics around that claim. The Tensorflow framework is built for the x86_64 architecture and Nvidia GPUs. There are no other builds for ARM or Apple M1 architecture.

Apple M1 CPU has one of the highest single-core scores at 1687; ok score on multi-core score at 7433 compared to the Intel i9–9880H (Macbook Pro 16”): 1029 single-core, 6012 multi-core score.

Looking at these values, it would probably work well for video processing, image processing, Apple Arcade, or gaming. The GPU will not work well for machine learning. It could, but just not the performance me and you are probably looking for. Something like the Razer Blade Pro 17 with the RTX 2080 Super would do better at 11.2 TFLOPS. You would be right with that thought. However, Apple also added in the M1 a Neural Engine integrated into the processor specifically for machine learning efficiency.

As for single-core score vs. the multi-core score, most data preprocessing libraries don’t take advantage of the multiple cores in the processor; therefore, the multi-core score does very little for data preprocessing. Having a better single-core score will do better in data preprocessing.