WWDC 2020: Apple Silicon
Reflections:
- First of all: it is going to be incredibly interesting to see where the architecture can go without the thickness of an iPad being a constraint. Shipping an existing iPad chip is basically as strong of a statement as they can make that they're not showing their cards yet.
- Making a big deal of virtualization still being there is necessary, but the way it was presented totally gave the (wrongful) impression that virtualizing Intel from Apple Silicon was possible. To the "UNIX-and-docker-using developers" that were mentioned, there's a hell of a difference between being able to use virtualization and containers for x86-64 vs for ARM, since the dry technical capability is intact but you miss out on the entire ecosystem of x86-64 containers and operating systems, which is most of the point. Were they trying to go through the keynote without using the word "ARM"?
- Considering that the transition will be about six months old as the first hardware is being shipped, I'm guessing the two year length of the transition will be necessary to develop hardware, architecture or OSes for the Mac Pro end of the spectrum.
- The unified memory architecture between the CPU and GPU is being touted and underlined as "modern" – I'm wondering where this leaves GPU support, even external. To a degree, even the Afterburner card seems dated by this framing, but maybe it'll be baked into the Mac Pro equivalent to begin with.
- If the unified memory architecture is such a big deal – will any Mac even be able to have user-installable memory after the fact? The term of art used during the presentation was SoC – System-on-Chip – and not CPU and GPU, and for them at least in the current form factors, all direct RAM usable by the processing units is hooked up inside the die. I guess they can carve out an iMac Pro/Mac Pro-sized exception to allow plebeian DIMM modules in addition to the on-chip RAM too. This session may contain answers.
- All demos were seemingly made on Pro XDR Displays, which use Thunderbolt 3 only for video signaling and not USB-C – if I'm not mistaken it runs two parallel DisplayPort streams to be able to fill up the display, leaving only enough pins and capability for USB 2.0 on its USB-C hub. But the tech specs of the Developer Transition Kit lists only USB-C, USB-A, Gigabit Ethernet, and a single HDMI 2.0 port. Did they all use some screwball converter to the HDMI port, which would have to be enlightened to their peculiar multiplexed Thunderbolt-DisplayPort connection?
- Update: Apple will continue to support Thunderbolt, with strong references to Apple and Intel having co-invented it, maybe to distance them from AMD which has had a famously hard time getting support. The statement doesn't include a reference to a version number, so it could be read as Thunderbolt 3 being supported by its inclusion in USB4.
- OpenGL support will be present-but-deprecated from the start, which essentially means the full OpenGL stack (beyond OpenGL ES) is available.
- Rosetta for PowerPC/Intel was barely able to run an Office+Photoshop demo convincingly and was labelled "fast (enough)" in the slides; with Rosetta 2, we saw recent-ish AAA games and 3D modeling software being labelled as "great" and "fluid" and "without hitches". Doing it up front surely helps, but they've raised the bar of expectations by a lot this time. If they're launching as soon as by the end of the year, they'll have to deliver.
- Being able to use XPC to support Intel and ARM plugins separately is inspired. I do wonder how many applications in the target audience allow for such a platform-specific architecture though.
- Depending on how things shake out especially with the desktops, this could be the end of Mac being a "PC"-family architecture. The screws will be put to anything new that has to be brought along, and many things are carried on only reluctantly and/or temporarily. The explicit mention to still be relevant to multiple-volume-multiboot-OS-external-drive-UNIX mavens is a strong signal they don't intend to go all the way, but whether it'll be enough for people who need something that's PC-like in its structure is anyone's guess.
- No word on whether apps from unidentified developers will still be allowed. (Update: allowed, but notarization is required.) The UNIX mention is interesting, because cross-platform command line tools can't really be expected to be fully packaged as macOS-enlightened, including notarization.
- Running iOS/iPad apps seems like a gimme and recontextualizes wanting to make Catalyst so badly, but also seems like even more of a half-solution without a touch screen, which does not seem likely without API to enable it, which would have been announced now. Then again, iPadOS pointer support sprung up virtually overnight.
- Having Office, Photoshop and Unity ported enough to be running from day one is far from "let's fly out the guy from Mathematica the week before". The Intel transition, even though the technical foundations had been laid for years, was famously close-hold; I wonder how long this has been cooking?
- Having prepared ports of open source components is also a sign of the times.