M1 Macs: Truth and Truthiness

Are you sure about Exynos? They announced the 1080 just a couple of weeks ago and the 2100 is expected next year.

By clever I mostly mean, from what I've seen from reports, handling stuff like transition from x86 to arm by baking in instruction translation into the die. Their approach with some things is a bit different from what others are doing, but it's not like they wholly invented a new 2nm process that breaks what we thought the laws of physics were, they're using the same fab as everybody else. Thus their design at the moment is clever.

I get the "fast and hot" v "slow and cold" history of devices, and am pleased to be told that the M1 has broken this. However, and it may well be that I haven't understood complexities here, a recent Ars review [1] (glowing, by the way) of the M1 MacBook Air, contained this passage:

According to Apple, the MacBook Air's M1 is voltage-limited in order to function within the fanless design's thermal envelope. iFixit's teardown shows in detail that the Air's M1 cooling setup is an entirely passive affair, with just a heat transfer plate in between the M1 CPU and the aluminum body. I was expecting performance similar to but perhaps a bit lower than the M1-powered Mac mini, and that's more or less what I got. However, the Air's M1 is good for at least a few solid minutes of full-bore Firestorm core performance before it throttles back.

In benchmarking, I noticed that subsequent runs of the Final Cut Pro export would slow down dramatically—the first export would complete in about 1 minute and 19 seconds, but if I immediately repeated the export it would take a bit under 2.5 minutes—and the Air would be quite warm to the touch. After closing the lid to hibernate until the Air was cool and then repeating the export, the time was once again in the 1:20-ish range.

To create some more sustained load, I cloned the source video three times and then repeated the export process. Starting from a cold startup with the MBA's chassis at ambient temperature gave a result of 4 minutes, 21 seconds. This time, I opened Activity Monitor's CPU graph to spy on the core utilization. All eight cores were engaged until about 2:56, at which time half of the cores—presumably the high-performance Firestorm cores—dropped to less than 50-percent usage and stayed there until the run completed.

A second run immediately after that took 7:37—not quite twice as long, but heading in that direction. Activity Monitor's CPU usage graph showed half of the cores (presumably the high-performance Firestorm cores) at half utilization for the entire run.

Further testing—including several runs after letting the MBA sit powered off for about an hour to make absolutely sure it was cooled to ambient—failed to produce anything resembling a precise, repeatable time interval for when throttling starts. The best I can do is to say that it seems that when you throw a heavy workload at the MBA, it runs at full-bore until the Firestorm cores become too toasty, which seems to take anywhere from 3-ish to 6-ish minutes. Then it backs the Firestorm cores off until they show about 50-percent utilization, and the amount of heat generated at that level seems to be within the sustained thermal capacity of the design.

My amateur reading of this was that, yes, the M1 is incredibly fast when cold, but when it warms up it becomes measurably slower because of throttling - even if still fast by comparison to previous Air models.

[1] https://arstechnica.com/gadgets/2020/11/apples-m1-macbook-ai...

Yeah, that would be really amazing. Is it even possible to install Linux on this new Mac though? Probably not. Will Apple offer these chips to other manufacturers? Would be cool if we could build small form factor PCs with these M1 chips!

I also wonder about the driver situation. Intel hardware has excellent support on Linux, everything works out of the box. Apple probably has zero interest in contributing drivers to the Linux kernel.

Qualcomm has been unable to compete with Apple's ARM phone processors for years. It's gotten to the point that the only processor that comes close to a new Apple SOC on benchmarks is Apple's previous SOC. Why do you expect the PC market will be different than the phone market?