Any Update on FFXIV Compatibility for ARM (M1) Macs?

[Shibi]

I recently purchased a Macbook Air (M1), and I was curious to see if FFXIV would run on it. I did not purchase the Mac download, as based on the footnote, I was concerned it would not run at all.

It certainly runs. I think it's a case of old web pages, or more likely slow certification by the support team. It took them ages to certify Big Sur.

[Packetdancer]

It certainly runs. I think it's a case of old web pages, or more likely slow certification by the support team. It took them ages to certify Big Sur.

I suspect it's more that at this point, the Mac version is really the Windows version, running atop an API translation layer (Codeweavers Crossover), which itself is running through Apple's Rosetta 2 (which allows Intel code to run on M1 Silicon). And the graphics are written in DirectX, which Crossover translates into Vulkan 1.1 using DXVK, which is then translated into Apple's underlying Metal graphics API using MoltenVK. So while it works, that's a lot of moving pieces—many of which are outside of Square-Enix's control. It's not too surprising to me that they'd be hesitant to claim official support, since if they do and something breaks, it may be out of their hands to fix (without writing a full native version of the application).

For instance, there was an issue early on where downloading a patch would cause the game to stop executing on M1 Macs entirely; it would just instantly crash when you ran the game, with no explanation. And there was basically nothing that Square-Enix could've done to fix that, because it had to do with how Rosetta 2 handles things.

As brief explanation, Rosetta 2 is not an emulator; it's actually an AOT compiler—AOT meaning "Ahead Of Time", where you compile a program targeting a specific architecture just once, and store the resulting native executable. (As opposed to Just In Time compilers, where you have a program stored in a more universal generic bytecode, and then compiled into native code when you run the program.) The first time you run an Intel program through Rosetta 2, it literally breaks down the program, analyzes it, and creates an entire ARM-native version of it, which is then cached. This means you don't take the emulation hit, since you're not running an emulator that has to pretend to be something else; you're literally running a native ARM executable compiled for the M1.

And since ARM is actually a more efficient architecture than the somewhat-aged-and-creaking x86 architecture, you can sometimes see some pretty substantial performance this way. As people have seen with the FFXIV client; the Rosetta-derived ARM version performs better on M1 silicon than the Intel version does on actual Intel Mac hardware. Which is frankly breathtakingly impressive.

However, Rosetta 2 needs to be able to tell that a program has changed to know to recompile it; as best I can tell, the way the x86 client was stuffed inside of Crossover's WINE-derived containers meant that when the game data changed in a patch but the Crossover portion did not, Rosetta 2 seemingly didn't know it needed to re-do the compile... which meant it tried executing an outdated ARM translation of things after a patch, causing an instant crash. You could work around this by fiddling around inside of Crossover so that Rosetta 2 knew to generate a new ARM version of it, but that's mucking about with things at a level that Square-Enix almost certainly would not want to try to provide support for; breaking open the .app bundle and fiddling around with the innards is a bit beyond the scope of what your average customer support person is prepared to talk folks through.

It is thus far easier for them to just say "this is not supported" and let people take the plunge themselves.

One hopes that while taking this approach, they're also trying to figure out what their official support strategy for M1 silicon will be in the long run—though I won't be exactly surprised if that strategy is "end-of-life the Mac client".

(Rosetta 2 has since added provisions to recognize scenarios like the Crossover one, mind you. So that particular problem went away back in like... late December or early January or something.)