Quote:
Originally Posted by Kushan
I think there's going to be a big distinction between Apple's ARM chips compared to the intel equivalent and Apple's ARM chips compared to previous Apple implementations of those chips.
This is something I watched recently and is related to the subject. Essentially, Apple's recent devices haven't been particularly great at thermal management of those chips. I think at one point in the video Linus speculates if it's deliberately bad so that when it comes to the ARM chip switch, it doesn't look like you're loosing much (any?) ground. While that's a throwaway comment and not meant to be taken seriously, I do think there's possibly a sliver of truth to it.
So you'll have this situation where you'll see comparable or even better performance than the previous gen Apple machines, but realistically the performance won't actually beat x86 if x86 has adequate cooling. In other words Apple will use murky comparisons to skew things in their favour. And maybe that's fine, because you can argue that it's a strength of ARM to require less cooling, but you can tell from the Video that Apple really half-arsed their cooling. |
The thermal advantage only exists if Apple don't cram too much stuff on the SOC, ie: Co-processors, encoding chips etc etc. Cramming more components onto a tiny surface area doesn't really bode well for dissipating heat.
We don't really know much at this point what is going into the new machines. The only information which seems to be out there are some supposed leaked benchmarks from the developer kits Apple have been sending out. This is probably the most informative article:
https://arstechnica.com/gadgets/2020...ransition-kit/
Main things to note:
- The developer kit is a Mac Mini running a hobbled A12X SOC from the iPad Pro (half of the cores disabled)
- Benchmarks are from non-native (x86) Geekbench tests via the Rosetta 2 interpreter
- Results: 20% slower benchmarks than an entry level i3 Macbook Air on single thread performance
- 38% faster for multi threading
Which is not too bad really if you factor in the fact there's a performance penalty associated with not running native ARM code.