Why doesn't Apple Silicon support Windows on Arm if it supports Linux on Arm?

[Homy]

Because Apple doesn't want its laptop batteries get drained.

 

[mr_roboto]

Probably the old naive "VMs slow", "native boot fast". Go figure.

I mean, it's not really naive. VMs do have overhead, and people notice it sometimes. Life got better at work after we stopped using Linux VMs as interactive workstations and went back to native Linux.

(*) I just recently spent some time on a GPU programming forum arguing with people who were very convinced that Vulkan is better because it uses fixed-layout structs and simple function calls instead of more abstract interfaces. Makes the API "faster", they said. As if saving a dozen or two of machine cycles on API call will make any difference at all if you have to do a user/kernel transition to hand the data off to the GPU... the lack of system programming education and understanding in the C++ crowd is frankly staggering. It's like most actually good programmers left for Rust or something...

Sounds like people who grew up on GL 1.x APIs where the overhead of each function call was super important because the data payload of each call was only one vertex (or was it one triangle? I don't remember), and haven't updated their thinking since.

 

[leman]

I mean, it's not really naive. VMs do have overhead, and people notice it sometimes. Life got better at work after we stopped using Linux VMs as interactive workstations and went back to native Linux.

With current hardware support for VMs, it doesn't have to be slow. I mean, GB5 running under Parallels on M1 delivers 95% of the non-virtualised performance and it can probably get even better with future improvements. The performance is more than good enough. What I was referring to are the "old good days" when every memory instruction had to go though the hypervisor...

Sounds like people who grew up on GL 1.x APIs where the overhead of each function call was super important because the data payload of each call was only one vertex (or was it one triangle? I don't remember), and haven't updated their thinking since.

Immediate mode GL has not been a thing for decades, I doubt that these people even knew about it. It's simply the crowd that would burn you on the stake for using a single indirect function call (because "slow") while at the same time telling you to create a bunch of std::vector in a performance-critical loop (because "C++ standard library is optimised")

 

[Gerdi]

With current hardware support for VMs, it doesn't have to be slow. I mean, GB5 running under Parallels on M1 delivers 95% of the non-virtualised performance and it can probably get even better with future improvements. The performance is more than good enough.

In general I can confirm this. When running GB5 on my Qualcomm 8CX Gen 3 device, the difference between native Linux and Windows WSL is below 1% - I am always getting scores between 6350 and 6400.

 

[MandiMac]

I downloaded ARM for windows for my MBP and I'm personally unimpressed. Apple has done a much better job with its Rosetta 2, then MSFT has with their emulation.

There's also MS' Dev Kit available for sale as well

As a follow-up, Ars Technica has that one covered as well: https://arstechnica.com/gadgets/202...icrosofts-600-arm-pc-that-almost-doesnt-suck/