Hacker News Reader

Comment by boromi | Hacker News Reader

why does arm64 need virtualization on apple silicon? Isn't M1 chip referenced on the Arm architecture?

It's virtualization, not emulation. It is indeed native arm, which is why virtualization suffices. To run x64 windows you would need to emulate.

The virtualization is needed because you are running the guest OS on top of the host OS. The same situation as if you were running x64 windows on an intel mac (e.g. via Virtualbox, vmware, or Parallels): in both cases you need kernel-level support in the host OS so that it can act as a hypervisor, and hardware support from the processor (in Intel it's done via vt-x) to support things like nested paging, ring -1, etc.

My knowledge may be out of date here, but I seem to recall that with a quality hypervisor (ESXi and even VMware workstation running on a host OS) you can see performance around 90-95% of bare metal. Of course a type 2 hypervisor will depend on what the host OS is doing, but if it's working at a 50% load then the guest OS should perform almost as though it had 50% available. (Yes, lots of variables will factor in here)

Not sure if we'll get there on an M1 without a heck of a lot more work, but it's early days and things show lots of promise.

What I really want though is not to have to own Apple hardware for this. The M1 shows such promise that it would be crazy for this sort of chip not to be pursued as the next level of commodity chips.

A significant reason for the high performance of Mac OS on M1 is that the M1 has been tweaked to optimize common Mac OS/ObjectiveC/Swift operations. In one interview, Apple engineers said that they worked on speeding up the operation to allocate and deallocate objects by 5x over Intel chips. That kind of operation is common and used in loops and in the reference count garbage collection used by those languages. By doing this kind of optimization they can tune the SOC to the OS. That is hard to do for commodity chips.

Given Windows on the consumer side & linux on the server side, is there any reason, (assuming difference chip design paths for server vs consumer) that similar optimizations couldn't become fairly standard on commodity chips? I'd think MS especially would be willing to have some tight partnerships with chip devs to do so.

Could you point the interview?

https://twitter.com/Catfish_Man/status/1326238434235568128?s...

I presume a lot of the perf gains of hypervisors are using Intel vt-x to hardware accelerate.

The other issue with ARM is that there are several dozen chip varieties whereas x86 only has Intel and AMD. This could hopefully accelerate mini-CPU companies springing up with their own ARM64 chips producing their own hardware running Windows ARM64. Google is rumored to be designing their own CPUs for future high-powered Chromebooks and Pixel phones. Qualcomm has an ARM64 chip powering the Microsoft Surface ProX but it’s nowhere near the performance levels of M1. Microsoft is starting to experiment with CPUs with the Pluton chip, maybe they will design an “M1” flagship chip going forward for their own Surface product line.

This could hopefully accelerate mini-CPU companies springing up with their own ARM64 chips producing their own hardware running Windows ARM64

I hadn't even thought of that. There's real potential to break open the market on general computing/desktop/laptop chips here. I can see how that might get complicated from an OS support standpoint, but it could also instill a remarkable amount of innovation-driven competition. It probably already has, with board-room meetings (and more importantly some long lunches with engineers hashing things out) realizing the M1 is the one to beat. And since it's Apple, and Apple will probably remain niche & not license their tech to others, the first to market with even an 80% M1 solution will be in good shape.

Honestly I'm most impressed by the GPU side of the M1. Rough equivalence to a dedicated 1050 Ti is crazy. That's in the respectable-low-end gaming side of things: I'm on a weird Intel-AMD hybrid chipset (hades canyon) that puts me between a 1050 and 1060 and I'm running AC:Valhalla on moderate settings ~45fps. It seems like Rosetta can achieve something around 70% performance, so less demanding new games and most older games should be playable for anyone who just wants to play and doesn't mind a steady 30fps and modest textures & AA. I think some Wine variant (CrossOver?) is already working to make native Windows games playable under Rosetta as well.

It wouldn't happen overnight, but the prospect of making AAA games playable on a $700 off the shelf mini of $1000 mac book would spur on the big game engine makers to port their tools over to make M1 a build target. Sure if you're tech savvy and shop around and don't mind at least a few hours of integration hell you can build yourself a very budget low-end gaming pc for ~$500 by forgoing a dedicate GPU and relying on the Vega 11 in Ryzan 5 3400G and that's without an OS and a much larger footprint. (Though if Steam gets their act together and offers a cloud-streaming option where I don't have to repurchase my games ::ahem, Stadia:: I'd forgo gaming hardware and stick to a strong Ryzen CPU w/ solid APU for non-gaming tasks.)

Is Microsoft planning on arm emulation of x64 as well?

Seems like a very interesting future if they are.

They already do:

https://www.theverge.com/2020/9/30/21495510/microsoft-window...

They have been doing for more then a year.

Reply on news.ycombinator.com