Tested: How badly Windows on Arm compares to the new Mac M1

All,

A small dose of schadenfreude for a rainy Wednesday morning.

This article is worth a read to see how Windows on Arm compares to macOS. (Spoiler- not well!)
It is also interesting that this is posted on a website dedicated to PCs!

Tested: How badly Windows on Arm compares to the new Mac M1s

Microsoft’s 64-bit X86 emulator is still in beta, so we can’t make definitive statements about its success. But it’s hard to believe that further development will bridge the vast gulf of performance between Windows on Arm and Apple’s M1-based Macs. In six months, Microsoft may be able to boast that its emulation performance has improved by a significant amount. But without the combined miracle of a much better CPU from Qualcomm or another Arm chipmaker and continued improvements from Microsoft, the future of Windows on Arm looks grim.

Thanks for posting that. IMO the article is unclearly written. I couldn't tell which Windows-on-ARM tests were emulated and which were native.

We know the M1 is really fast and so is Rosetta2. Apparently the Qualcomm ARM CPUs are much slower, as are the SQ1 and SQ2 ARM CPUs co-designed by Microsoft. It seems the Windows ARM emulator for x86 code is also slower than Rosetta2. It makes sense that even native Windows-on-ARM benchmarks would be slower than native Apple Silicon benchmarks. But which Windows benchmarks were native ARM binaries and which were emulated x86 binaries? The article is very unclear on that.

Apple's goal (for both themselves and Mac app developers) is rapidly port all x86 MacOS apps to Apple Silicon. In the interim Rosetta2 will provide fairly good performance of emulated x86 apps. But the end point is clearly defined -- it's a rapid, complete hardware and software transition, with major benefits.

Microsoft's strategy seems less clear. Will there be ARM desktops running a large portfolio of recompiled ARM apps on Windows? Will there be an ARM version of SQL Server running on a rack-mounted ARM server? Instead of a transition, it appears Microsoft will belatedly support some apps natively on lower-end ARM devices. 

It is comical that only six months ago, there were projections that Apple could be in for a "rough ride" on the ARM transition. Former Microsoft VP Steven Sinofsky basically felt that because MSFT had trouble with their failed Windows RT project, Apple would have similar results. See below Sinofsky article from June 10.

The Apple picture will become clearer when the first desktop Apple Silicon Macs ship next year, and when a broader range of desktop apps are ready. But it appears within a few months there will be a 16-core iMac running native Apple Silicon versions of released, supported Adobe and Microsoft Office suites. It's not a stretch to imagine it will be several times faster than the current M1 Mac which only uses four "power" cores (and even that is pretty fast).

Sinofsky and some of the other commentators may have gotten backwards who should teach whom about how to conduct a CPU transition. 

www.theverge.com/2020/6/10/21285866/mac-...s-transition-coexist

medium.learningbyshipping.com/apple-maci...cessors-bff8d191a7ff

One very significant aspect of the Rosetta 2 emulation/conversion that even Apple were slack in emphasising, is the fact that a large proportion of the Intel to ARM translation is done at the time an app is installed. As a line on a caption during the keynote says, "Translated at install time", but this significant fact was glossed over! The original Rosetta for PPC to Intel emulation didn't do this, but in my experience the overhead was not deal-breaking by any means. Rosetta 2 sounds virtually invisible.

There's a very good article that explains how this is handled and why Rosetta is almost unnoticeable, here:

beebom.com/apples-rosetta-2-vs-windows-x86-emulation-explained/

There's a persistent impression given in lots of articles that the emulator is there in the background translating everything on the fly, so this cripples the speed. This is not the case.