As far as I understand, they don't emulate the x86 instruction set, but instead cross-compile it to ARM native instructions. That means that whatever implementation details are behind the patented x86 instructions are irrelevant, since there is no emulation of those instructions, they are just replaced by equivalent ARM instructions. Intel will probably still sue, but I expect them to settle out of court.
"The real-time ‘Just-In-Time’ transcoding emulation that converts x86 instructions to ARM is done the first time you run the software, and then it’s cached by Windows"
Still sounds like emulation to me. It's a pretty fine line.
Isn't that equivalent to emulating it? Like emulators for video game systems that have both an interpreter and a JIT recompiler, the latter is just an optimization of the former. They both must interpret the instruction set of the target and run equivalent instructions on the host.
Patents have a funny feature, in that they (are supposed to) apply to a specific implementation. So "equivalent effect" is not enough for patent violation in the general case.
No, but they can do things like not optimize code paths and not work with Intel on power saving features. The average consumer is far more likely to buy a Windows laptop with an arm cpu than an Intel laptop running a different os. Intel won't win that fight unless it's in a court room.
It could cut support for the highest end x86 CPU's in desktop machines (ie. non-server). That would then make those CPU's un-sellable, since they're pretty much all used for gaming.
Obviously, they wouldn't say that - the way to market that is to release 'Windows Gamers edition', which costs way more and 'unlocks' the power in the latest CPU's. When in fact, non 'game edition' windows versions simply leave some cores disabled or sleep some of the time when they detect a 'too fast' CPU and a fullscreen game running...
> At some point I'd like to believe people would just stop paying Microsoft tax to play computer games...
I wish I could. The games I like aren't released anywhere else.
> Then I remember people pay tons of cash for their rigs that are obsolete within two years anyway.
This is false. A moderate gaming PC can last longer than a console. I used my previous rig for 4 years and now my wife is using it perfectly fine. It still has better visual fidelity than a console.
Expensive gaming rigs - a bit less so than PC, but for equivalent hardware, about $400 cheaper at introduction, about equal once 1-2 years pass.
Obsolete in 2 years - check
Expensive software - much more so for consoles.
So right now the situation is that if you go for PC gaming instead of console, you'll have about the same expense, but 100 games for PC (and a powerful general purpose computer), and maybe 5-10 on console.
Almost certainly. There may be small corner cases where the translated code is faster but I’m sure they’re the exception.
That said, people may not care. ARM can be very fast. My iPad is faster than many low end laptops.
So you get in the situation where a low cost laptop with ARM may be as fast or faster going through emulation than a low end Intel chip from 4-6 years ago while providing WAY more battery life.
People may be willing to make that trade.
I’m guessing it’s not 20% speed on most tasks. Probably 80%+, depending on the age and power of the Intel chip.
I think 20% is more likely than 80% for raw compute tasks - ARMs are competitive with intel in performance/watt, but are not much better at it (and definitely are not competitive yet in absolute performance). People are going to buy it for the 20-hour battery, and then run the CPU at 100% to run emulation which will eat the battery in 3 hours.
However, most people run a small selection of programs most of the time (web browser, word, excel, outlook), which will likely have an ARM native version sooner than later, and the long tail of programs they must have are often not compute intensive, and spend 85% of their time in the OS doing graphics/io which are native. So overall, even if the CPU emulation is 20% speed, the overall feel is likely going to be 80% or better for most people.
I don't see any mention of x86_64/AMD64 in the article. I would expect a version of Windows which can't run x86_64 apps to result in a lot of nasty compatibility surprises--it's been awhile since developers had to support 32-bit-only hardware--so I'd expect Microsoft to support it if at all possible.
You're right, the posted Anandtech article does not mention x86_64; however, other prior and recent coverage of this topic does confirm that x86_64 is not supported, such as this Ars Technica article [1] from the same day.
There's no support for x86_64 binaries. Windows 10 is still available in 32-bit editions and almost all apps are still available for 32-bit, so I don't think this is a big issue.
I assume any kernel drivers will need to be compiled for ARM64 though. This might not help corporate adoption--VPN client, 3rd party anti-virus/full disk encryption, etc.
Also one thing that is x86-64 only is WSL, the Windows Subsystem for Linux.
I think most Windows applications which are truly 64-bit only are either too large to run on a machine like this, or likely to be rebuilt for AArch64. For the time being, I think IA-32 support is enough to tide them over, and it has the benefit of being thoroughly out of patent (except some fairly common extensions, like SSE2, which will expire soon).
There's still a lot of Windows devices that run x86 Windows, and new devices still ship with it, mostly low-memory or low-disk-space devices. Thus apps need to support it.
The Intel patents for up to SSE2 should expire soon, and hopefully the AMD64 patents should be licenseable from AMD (if they are interested in that). I would not be surprised if this already emulates up to SSE2, given the likely release date and the fact that recent versions of MSVC already enable /arch:SSE2 by default.
> I would not be surprised if this already emulates up to SSE2, given the likely release date.
Yeah, even if they could be sued, they could settle for a couple years damages max, or quite possibly win, since they're doing DBT (no different from QEMU or loads of other x86 + SSE2 emulators).
though it's worth remembering that QEMU developers do not have the deep pockets that MS has -- and it is the depth of the pockets, rather than the (alleged) violation that drives law suits and threats.
My most important difference is whether Microsoft is going to stop supporting it a couple years after selling it like they did with Windows RT--leaving their customers with thousands of dollars of useless, unsupported machines.
I still can't believe they did that. Discontinuing software is one thing--you can find replacements. But discontinuing Surface RT tablets OS right after selling them is insulting and I will not be recommending them to my clients until Microsoft proves it will actually provide a long-term platform.
