Windows 10 on ARM Support?

[AskrHinnRauthi1848]

It seems that new computers are coming out that are going to have way better battery life and run on ARM procesors [www.theverge.com] . It would be great to know that Steam will have support from Day 1 for these computers to play CS:Go and Civ VI.

I plan to get one to replace my Lenovo T420. Knowing tha Steam will run and that applications will be updated for ARM support will be great!

[Dr.Shadowds 🐉]

Messaggio originale di MossyRathalos:

Messaggio originale di x_wing:

Steamlink has an arm cpu inside so yes, they can handle streaming. In fact, a dude of my work fund an kickstarter and got an arm notebook (he installed Arch Linux on it). The SoC was able to reproduce 4k video and handle multimedia a web browsing without problems.

Maybe ARM CPUs are far away from Intel top end, but I'm confident that a high end SoC can be superior than any low end system with pentium or celeron CPUs.

You do realize that games almost never support the ARM instruction set, correct? The only things you could play are mobile ports and W10 Minecraft, as they support the instruction set.

It's not about who develops the processor, it's about what the processor does.

^This.

[x_wing]

Messaggio originale di MossyRathalos:

You do realize that games almost never support the ARM instruction set, correct? The only things you could play are mobile ports and W10 Minecraft, as they support the instruction set.

My answer was focused to show my experience on ARM SoC and to mention that this CPUs are pretty decent for daily use on a desktop environment, but I never mention that they would run any
nowday game crosscompiled (in fact, W10 RT emulation layer is a bad idea IMHO).

Messaggio originale di MossyRathalos:

It's not about who develops the processor, it's about what the processor does.

Don't know what you mean with this, but do you think that an Apple A11 is the same as any MediaTek SoC?

[Zaquilhões]

ARM= low power architecture designed for ultra low consuption phone/netbooks,dont expect nothing fancy,steam and devs probably wont even care,also there is ports of ARM of old games to android,like GTA SA,max payne,etc.

[Davo SVK]

♥♥♥♥♥♥♥♥. Everybody is talking about how has ARM terrible performance. BUT ITS NOT TRUE
These new Windows on ARM devices have SnapDragon 835 which is much more powerfull than most x64 CPUs in laptops. And Even more powerfull SD845 is comming soon

[vadim]

Messaggio originale di Davo SVK:

which is much more powerfull than most x64 CPUs in laptops. And Even more powerfull SD845 is comming soon

That isn't true. And cannot be. ARM CPUs have relatively the same IPC as Atom and the same (or even lower) clock rate. Thus, they obviously much LESS powerful than even lowest grade x86.

[aiusepsi]

Messaggio originale di Kokain:

ARM= low power architecture designed for ultra low consuption phone/netbooks,dont expect nothing fancy,steam and devs probably wont even care

Not so; the ARM architecture was originally developed for desktop machines back in the 80s; it's actually sort of an accident that it happened to work really well for low-power chips.

Messaggio originale di vadim:

That isn't true. And cannot be. ARM CPUs have relatively the same IPC as Atom and the same (or even lower) clock rate. Thus, they obviously much LESS powerful than even lowest grade x86.

Not so! Have a look at the Geekbench scores for the Apple A11: http://browser.geekbench.com/ios-benchmarks The A11 is scoring 4200ish, and the score is calibrated such that 4000 is the score of an Intel Core i7-6600U.

Admittedly, Apple's ARM processors are pretty far ahead of their competition (which is one of the reasons it's weird that there are Windows ARM laptops before Apple came out with ARM Macs) and the latest-and-greatest x86_64 processors are faster, but the gap is not so vast as you're making out.

I don't think it's likely that you're going to want to play the most intensive games on an ARM laptop any more than you'd want to play them on an x86_64 laptop, but there are plenty of games out there, especially older games, which would be just fine on ARM. It'd be good if Steam could at least do what they do for 64-bit support, i.e. allow a game to download different binaries depending on the CPU type, and to launch the appropriate executable, so that if games do want to support ARM natively (and for many games, this could be as easy as a recompile! For the others, they might want to rewrite stuff that's using x86_64 SIMD intrinsics to use ARM ones) they can.

[vadim]

Messaggio originale di aiusepsi:

Not so! Have a look at the Geekbench scores for the Apple A11: http://browser.geekbench.com/ios-benchmarks The A11 is scoring 4200ish, and the score is calibrated such that 4000 is the score of an Intel Core i7-6600U.

Sorry, but this is common misconception. Apple A11 is a very specific chip that has additional coprocessors. Namely, cryptographic. While one of the main tests performed by geekbench is just encryption.
it shows coprocessor performance, not CPU (you can see it scales fairly bad in multicore test). This is only one of the reasons, of course.

[aiusepsi]

Messaggio originale di vadim:

Sorry, but this is common misconception. Apple A11 is a very specific chip that has additional coprocessors. Namely, cryptographic. While one of the main tests performed by geekbench is just encryption.
it shows coprocessor performance, not CPU (you can see it scales fairly bad in multicore test). This is only one of the reasons, of course.

It's not that specific; SHA and crypto instructions are part of the Aarch64 instruction set, so any 64-bit ARM CPU will have them. Coprocessor is not really the right term; they're part of the core in the same way as, say, the floating point unit is. The question is, is the balance of floating point, crypto instrinsic, integer operations, etc. representative of a typical workload. Apparently, the crypto workload of Geekbench 4 only counts towards about 5% of the score (page 6[www.geekbench.com]) so it wouldn't seem to be over-weighted to me.

It should also be pointed out that there are AES instructions for x86 (in most CPUs from Sandy Bridge up), and there are SHA instructions shipping in AMD's Ryzen processors, so accelerated crypto is not unique to ARM.

The multiplier on A11 for multicore is 2.4, which is not too surprising given that it only has two high-performance cores; the remaining 4 are low-power cores, and you almost never get a multiplier equal to the number of cores anyway (c.f. Amdahl's law).

[vadim]

ARM is a typical RISС CPU. It has only two instructions that can work with memory arguments - load and store. To use immediate operand, you may need to write several commands. Each of which will have only part of this operand.
The set of SIMD commands (NEON) is significantly inferior even to SSE, not to mention AVX. For example, this processor does not have a division operation. At all. Neither integer, nor floating, nor vector. And so on.
Thus, x86 can perform roughly the same number of instructions per sec as ARM, but these are complex instructions. ARM code has about 2 times less code density (I spend enough time doing cross-compilation of Linux applications).

[aiusepsi]

Messaggio originale di vadim:

ARM is a typical RISС CPU.

You'd hope so, otherwise they'd have to change the name. Seriously though, the difference isn't that huge these days; they're both decoding instructions into microcode on the front end.

For example, this processor does not have a division operation. At all. Neither integer, nor floating, nor vector. And so on.

64-bit ARM does have a division instruction. (Section 5.5.2, Page 47[www.element14.com]) There are plenty of ARM CPUs out there which don't have a divide, but the high-performance 64-bit ones do.

Apparently, the worst-case latency for an SDIV on a Cortex-A72 is 12 cycles (Page 9[infocenter.arm.com]), on a Intel Broadwell, it's 36 cycles.[gmplib.org]

As a general rule, more complex instructions take longer to perform. This isn't so bad because the CPU can out-of-order execute and perform speculative execution past branches (probably a big part of the reason for the A11 being so fast; it has a pretty deep pipeline compared to other designs). So there's not so big a win either way, really.

ARM code has about 2 times less code density (I spend enough time doing cross-compilation of Linux applications).

This is a reasonable point in that fitting more instructions into a smaller space means you spend less time fetching instructions from main memory, and more instructions fit in L1 cache. This is all good. It's a part of the reason for the existence of the Thumb mode.

I also spend a lot of time cross-compiling (or, right now, writing tools to automate cross-compiling). I'm cross-compiling the same codebase for x86-64 Linux, macOS, FreeBSD, and iOS Simulator, and ARMv8 & ARMv7-a for Android and iOS. The delta between size of code isn't all that big between architectures, although because of difference in the way they're linked they're not directly comparable.

Anyway: the point is: yeah, ARM is not so much slower that playing games on one is inconceivable. I mean, you would think that games like (Game Awards 2017 GOTY) Breath of the Wild or Super Mario Odyssey would be proof enough of that.

[vadim]

Messaggio originale di aiusepsi:

64-bit ARM does have a division instruction. (Section 5.5.2, Page 47[www.element14.com]) There are plenty of ARM CPUs out there which don't have a divide, but the high-performance 64-bit ones do.

Thanks for the info. Never knew that. To be honest, I know relatively well ARM ISA up to v7, but when v8 has been released I already ceased my work with ARM ports and only read a bit about it (new regs, advanced thumb and so on).
So, I ready to believe that you know this arch better.

[76561198416683070]

Messaggio originale di CharlestONE:

So most existing Windows programs won't work on ARM running full Win 10 without being updated?

[deusexaethera]

I'd really like an update on this topic for 2021 now that Windows on ARM supports x86-64 emulation, and now that Apple is planning to switch their entire product line to ARM. I've been able to run Civ 5 and other 3D strategy games on my Samsung Galaxy Book S LTE (powered by a Qualcomm 8cx) with pretty good quality, despite not having a dedicated graphics card, which is similar to the results people have been getting on the new ARM-powered Macs.

Valve needs to prioritize ARM64 support, both for Windows and MacOS, in light of recent developments.

[Omega]

Messaggio originale di deusexaethera:

I'd really like an update on this topic for 2021 now that Windows on ARM supports x86-64 emulation, and now that Apple is planning to switch their entire product line to ARM. I've been able to run Civ 5 and other 3D strategy games on my Samsung Galaxy Book S LTE (powered by a Qualcomm 8cx) with pretty good quality, despite not having a dedicated graphics card, which is similar to the results people have been getting on the new ARM-powered Macs.

Valve needs to prioritize ARM64 support, both for Windows and MacOS, in light of recent developments.

ARM or another RISC instruction set will take over the home computer market in the next ~10 years.

x86 is dead.

With Valve you never know, if they have someone internally with an interest in this we could see an ARM build of Steam for Windows/Linux/MAC tomorrow. And if nobody internally cares that much or they just don't have the time it might be a few more years when RISC really becomes mainstream.


I'd like to interject that what you are refering to as ARM64 is in fact AArch64,

[Dr.Shadowds 🐉]

Messaggio originale di deusexaethera:

I'd really like an update on this topic for 2021 now that Windows on ARM supports x86-64 emulation, and now that Apple is planning to switch their entire product line to ARM. I've been able to run Civ 5 and other 3D strategy games on my Samsung Galaxy Book S LTE (powered by a Qualcomm 8cx) with pretty good quality, despite not having a dedicated graphics card, which is similar to the results people have been getting on the new ARM-powered Macs.

Valve needs to prioritize ARM64 support, both for Windows and MacOS, in light of recent developments.

It's up to them if they want to, maybe it happen in ~10+ years. But Mac users has been less than 5% for near decade, and that most 90%+ has always been Windows OS. ATM sitting at 96% is Windows where remaining is split between Mac, and Linux.

You have to rely on Rosetta from Apple, as that your x86-x64 support emulator to running your non-ARM base apps, and games.