Metal is a proprietary 3D technology ported from iOS. Apple chose to do another closed tech instead of adopting Vulkan or fixing their slow OpenGL implementation.

If you look at the numbers, you'll see that Metal is about 30% faster than OpenGL on OS X, which is interesting because OpenGL is about 30% slower on OS X than on Windows. Which means they're forcing developers to use their proprietary API to get comparable 3D performance. How is that a good thing?

To be fair (regarding "closed tech"), Apple was hardly the only one to release a propriety Vulkan alternative (there's also Mantle and DX12) and Apple released Metal to developers almost two years before Vulkan 1.0 was released.

The difference is the Windows driver model allows IHVs to implement Vulkan alongside the proprietary DX12. Microsoft may not support it but they've provided the framework to let it happen anyway.

In contrast Apple exclusively controls the macOS graphics stack top to bottom, so as long as they don't explicitly support Vulkan they're forcing developers to use their proprietary API.

Vulkan is based on Mantle, and DX12 can coexist with Vulkan because Microsoft doesn't take responsibility for all of Windows' OpenGL and Vulkan support. In this area at least, Apple really is doubling down on proprietary technology where others aren't so much.

One aspect of their decision to run with Metal that surprises me, is that they decided the cost of developing and maintaining an API of that level of complexity was worth it. I'd like to know what they were thinking.

I think they looked at how 'well' third parties developed Mac OS X drivers for OpenGL, and figured that wouldn't get better.

Assuming they knew about Vulkan in time (Vulkan was announced at SigGraph 2014, about a year before Metal's _release_, so that may have been tight), I guess they assumed things wouldn't get better if they replaced OpenGL with another open technology.

Also, from what I read, Vulkan doesn't do anything implicitly, OpenGL does lots of things implicitly. They may have thought that the optimum API lies between the two.

Don't Apple make their own crappy OpenGL drivers, not third parties?

I don't think Apple ever publicly stated anything about it, but my $0.02 is on "The latter seems more plausible to me" in http://apple.stackexchange.com/a/213893:

"There is debate over whether Apple make the entire driver set themselves, or just take NVidia/AMD's core code & add their own hardware/OS-specific code to it. The latter seems more plausible to me, but either is just speculation."

I also notice that that page mentions drivers that can be downloaded directly from Nvidia, but gives no indication that either driver is significantly better than the other. That is not inconsistent with them essentially using the same source code.

You know, I really think this line of argument on HN is getting old and needs to be relegated to the dustbin of intellectual discussion.

Saying "Well, but X did $similarthing" as a loose form of justification of whatever Z is doing, like some strange variation of the "not as bad" fallacy. (Though I understand you're point about industry modus operandi)

a cookie if anyone knows the actual fallacy this falls under

It falls under the appeal to hypocrisy fallacy, albeit with inverted logic.

I've seen it called "whataboutism"

"two wrongs don't make a right"

Basically making it like OSX's implementation of EFI vs. UEFI, or their Lightning cables vs. USB-C:

1. Apple creates their own implementation of something and pushes the industry forward;

2. Everyone else realizes it's a good idea, and standardizes on something slightly different/incompatible (usually with Apple on the standards committee!);

3. Apple continue using their own thing that Works For Them rather than adopting the standard.

That's incorrect with respect to Metal, AMD's Mantle predates Metal and is what pushed the industry forwards.

The closest Apple ever got to making a standard was 1394, but even then, other companies were behind it (and thus the famous story about Jobs wanting to charge for 1394 until he found out Apple would be paying a fortune to other companies).

EFI was made by Intel and UEFI was basically EFI 2.0.

Thunderbolt was made by intel (lightning was a cheaper spec from AMD that didn't require expensive active cables). USB-mini breaks and displayport mini is big. What we got instead was USB-C which unifies both of them. Thunderbolt used displayport mini, but the latest 3.0 spec uses USB-C. The fact that all three technologies are starting to unify on the same connector is very beneficial and none of this is dislike for Apple.

Likewise, AMD made the mantle APIs for lower-level work and then turned them over to Kronos where they became the basis of Vulcan. Apple's metal API and doubling down by leaving the Vulcan work group is either "not invented here" or vendor lock-in.

Hmm, what does "to be fair" mean in this context? That it's ok because other people did a shitty move as well?

It feels like you're completely ignoring the second part of his sentence.

Mantle is not proprietary. Metal being usable before Vulkan doesn't exactly make up for the complete lack of Vulkan support and anything past OpenGL 4.1.

Didn't Mantle give rise to Vulkan?

Yes, Mantle was donated to Khronos in early 2015 (or maybe late 2014) and became the basis for the new glNext effort (after the failure of Long Peaks)

Metal came out before Vulkan did, not sure you can couch that as them choosing to do another closed tech.

How is Metal any different from DirectX/Direct3D in this regard?

It's not.

Metal is basically their proprietary version of Vulkan isn't it?

Metal came out before Vulkan.

It's conceptually similar, but is probably more properly related to Mantle (the AMD API later donated to Khronos, which they developed into Vulkan) and its initial release predates that of Vulkan by ~18 months (Metal was announced in September 2014 for iOS, the OSX port was released in June 2015, the initial release of Vulkan was in February 2016).

Somewhat, but with more (forced) automatic resource management.