In November 2020, I bought the bottom end 8Gb/256Gb 7GPU MacBook Air to replace my much more expensive 16Gb/512Gb Intel base MacBook Pro. My rationale was that I would minimize the cost during a rapid transition: I was hoping that Apple would release an ARM-based MacBook with more cores and a better (mircro LED) screen in 2021. I still think this might happen, and you might consider if you can defer your purchase a bit.
A couple thoughts:
- Apple charges a lot for more RAM and SSDs. Since the RAM is soldered, make sure you purchase what you need. With regards to SSDs, I purchased an external Sabrent Nano - this is a much better value and I can keep it when I upgrade to my next computer.
- Part of getting a laptop is the small size. The Apple power chargers do not use GaN technology, so they are really large. I am very happy with the 30w Anker Nano II, which is a fraction of the size of the Apple unit yet supplies enough current to fast charge the laptop.
- The Air an overwhelming benefit over the Pro: physical function keys. Having come from a MacBook Pro with the Touch Bar, I can tell you I find the physical keys much better. Muscle memory, no worry about keys disappearing to save power, etc. If the Touch Bar was the standard interface and someone invented physical function keys, I am convinced they would have been awarded a Noble prize.
- The Air has no fan, it is truly silent. In my usage, it never gets warm.
- The M1 Macs keyboards (Air and Pro) are a huge improvement of the butterfly keyboards used by some previous MacBooks.
- While benchmarks show that the Air will throttle a bit under sustained load, most of my tasks do not seem to be impacted, real world loads that are limited by Amdahl's law (e.g. only spending part of their time threading) and waiting for data from the SSD means that the real world performance is pretty good.
You can see that the Apple M1 does really well on a wide range of loads:
https://github.com/neurolabusc/AppleSiliconForNeuroimagingGeneral points about the M1:
1. Rosetta works really well. Very few tools used AVX, and so you can run virtually every application. The M1 is fast enough that x86 software is generally comparable or faster than the same software running natively on a Intel MacBook.
2. I did find a few edge cases that may impact you with specific applications. OpenGL is simulated through Metal, and generally works outstandingly well, however Metal has no concept of a geometry shader, so if your tools use those they will not work (my sense is that geometry shaders proved inefficient in general, and are exceptionally rare). While Python and Numpy are native M1, the SIMD routines are not ported, this means for some tasks Rosetta is vastly faster than native M1 code, where on other tasks the reverse is true. This means on mixed tasks, there is not a better option, and it is obvious the native code has untapped potential.