New Apple Silicon Macs: MacBook Air, MacBook Pro, and Mac Mini

trs96 said:

What I'm interested to know is how did Apple get the single core performance to higher levels (430+ points) more than even the latest Intel Core i9 CPUs ? Has anyone read or seen anything about this ? I know that the M1 clock speeds are 400 MHz lower so what is different about M1 that produces these results ? My guess is that it's the SoC arrangement / layout with the ram in close proximity as well as the faster L2/L3 cache that makes data move faster at the lower clock speeds.

Click to expand...

Way above my knowledge level but listened to a discussion on accidental tech podcast that I think is relevant. I think the info was from some technical interview apple had given about the new chips.

Short form: different instructions take different numbers of clock cycles, and apple seems to have optimised heavily for common instructions. Do that and you get much, much more from each clock cycle.

Longer form: the example was some instruction (fetching memory for example or some simple operation) that took eg 5 cycles on intel and eg 2 cycles on apple's ARM. The operation is exceedingly common and hence outsized gains from getting this right.

That doesn't mean the ram/cache proximity isn't a factor, but that apple having the compiler, the chip design, APIs and a lot of the core operating system in-house meant they could optimise really, really well. Obviously a lot more to it like predictive branching and blah blah blah as well.

(I seem to remember this was one of the originally-touted ideas behind RISC was to have fewer instructions that are used all the time and could be super-optimised).

Honestly way above my knowledge level but I still found interesting.

(Podcast worth listening to and fun, not too technical)

Interesting-

With 100 Chrome tabs open the MBP 13" w/ M1 doesn't even ramp up the fans. Wow.

trs96 said:

What I'm interested to know is how did Apple get the single core performance to higher levels (430+ points) more than even the latest Intel Core i9 CPUs ? Has anyone read or seen anything about this ? My guess is that it's the SoC arrangement / layout with the ram in close proximity as well as the faster L2/L3 cache that makes data move faster at the lower clock speeds.

Click to expand...

The RAM performs similarly to a high-end dual channel 4266 MHz kit, albeit it seems at higher latency. In my view, the on-package placement has more to do with energy efficiency and space/component/cost optimization, than with performance. The big performance driver is the core architecture which is quite something really.

From what I've read, I'll try to offer 4 explanation levels of increasing nerdiness:

ELI5:

TL;DR:

Tech jargon palooza:

• Extremely wide architecture with an 8-wide decode block (ARM ISA facilitates this). By far, the widest of commercial CPUs (Intel is 5-wide and AMD is 4-wide).
• Immensely deep Re-order Buffer (~630 instructions), allowing for efficient out-of-order execution and high Instruction level-parallelism (Intel's is 352 deep, AMD's is 256 deep).
• A very large number of execution units. The Floating-point operations/cycle throughput is 4 times higher than Intel's and 2 times higher than AMD's. A single core can almost consume the total memory bandwidth (68.25GB/s LPDDR4X-4266) and saturate the memory controllers, which is something unprecedented.
• Very deep load/store memory transaction queues for the memory subsystem (~150 and ~106 deep respectively). AMD is at 44/64 and Intel at 128/72
• Massive and extremely fast L1 caches (3-cycle 192KB caches for data and instructions). AMD has 4-cycle 32KB, Intel 5-cycle 48KB.
• Huge and fast 12MB L2 cache.
• A large 16MB SLC cache, accessible by all IP blocks on the chip.

Unadulterated source:

We've been talking about the effect on hackintosh.

But you have to wonder what the effect is going to be on sales of the remaining intel macs. Particularly at the relatively high-end, or at least steps above the basic iMacs.

This is basically a version of the Osborne effect: introduce a barnstormer set of (relatively) low-end machines, and basically announce everything else above it will be replaced in the next 24 months.

For anyone paying attention, would you buy a more expensive MacBook Pro, a more expensive iMac, or even an iMac Pro? You've got to look at those numbers and just say "maybe I'll just wait."

I guess Apple can hope they'll buy a MacBook Air or Mini to tide them over until the one they really want is available, and then buy another - that's way more rational than plunking down $2000+ for an Intel Macbook Pro.

Of course it won't apply to some group of people who really need a powerful machine NOW, but otherwise, I'd really expect to see sales drop off (or some really significant price cutting).

I'm not rushing to a store but fully expect I'll buy a Macbook Air pretty soon.

This is frankly just so cool and so mind-blowing..

I will absolutely be moving to an Apple Silicon Mac sometime in the next year.

What I can't decide now is what to do with my existing build. I think I might just sell the whole thing (or part it out) and build a new gaming machine at some point.

Hack has been wonderful for this last decade and the 3 machines I've built...but for me, it's basically over at this point.

These performance numbers (from ENTRY level new Macs) are simply breathtaking.

Armoured said:

But you have to wonder what the effect is going to be on sales of the remaining intel macs.

Click to expand...

Disaster for anyone holding inventory, specifically of versions that just got the Apple Chip treatment.

I think there are going to be some mega sales on early 2020 Intel MacBookAir's for instance.

If I owned a relatively recent Intel Mac, I'd be fire-saleing it ASAP and getting value while I still could

Oh well, it seems end of Hackintosh for sure. May be couple of more years till Apple and most apps on macOS completely drop the x86 architecture. I will probably then at that point format the computer and install Windows 7 or 8 on it. In don't like windows 10 junk UI and all the telementary that runs behind that you could never turn off.

Armoured said:

(I seem to remember this was one of the originally-touted ideas behind RISC was to have fewer instructions that are used all the time and could be super-optimised).

Click to expand...

Yes, although RISC actually may have more more instructions than X64 (depends how their counted) they do simpler things (so more of them may be needed perform a particular task) - not so much a smaller instruction set as reduced instruction complexity.

The first pictures I've seen online of the SoC in the new M1 Mac mini. It's under the silver heat spreader. Adjacent to that is the ram. iFixit will do a much more complete teardown and description in the next few days.

Looks like they are using a good amount of thermal paste. Maybe even better quality than what they used in the 2018 mini. Many people demonstrated they could lower CPU temps by replacing that TIM ('18 mini) with something better.



Picture with the thermal paste removed.





This is the surprisingly large heatsink. Looks quite substantial for such a small Mac. It even covers the two ram modules to keep those cool too.