AMD and Nvidia Will Make Arm-based Processors for PCs

[LAwLz]

Summary

According to a report from Reuter, both AMD and Nvidia will offer Arm-based processors aimed at Windows in 2025.

It has long been speculated that Qualcomm has an exclusivity deal with Microsoft, so that only they are allowed to make Arm processors that work with Windows. According to several reports, that exclusivity deal ends in 2024. If that is the case, we should expect other companies to start releasing Arm-based processors for Windows in 2025. AMD and Nvidia are two of them rumored to be releasing that.

 

Quotes

Quote

Nvidia has quietly begun designing central processing units (CPUs) that would run Microsoft’s Windows operating system and use technology from Arm Holdings, two people familiar with the matter told Reuters.

-snip-

Advanced Micro Devices also plans to make chips for PCs with Arm technology, according to two people familiar with the matter.

 

My thoughts

It sounds like 2025 will be an interesting year for processors. I really hope Windows on Arm takes off, because if it does then we might suddenly go from 2 processor brands fighting each other, to ~5 brands fighting. That will hopefully drive more innovation and lower prices. Qualcomm recently announced their Snapdragon Legion X processor for Windows. This report says AMD and Nvidia are interested, and last year we had MediaTek say they were interested in developing processors for Windows PCs.

 

Sources

https://www.reuters.com/technology/nvidia-make-arm-based-pc-chips-major-new-challenge-intel-2023-10-23/

[Eigenvektor]

The Snapdragon X Elite sounds very promising, at least on paper. But I think the real obstacle to adoption of Windows for ARM is software support, not just raw CPU performance or power efficiency.

 

Unless the software people need/want works on an ARM based machines, I'm not sure it'll get much broader uptake than it has right now. Plus the amount of legacy software on Windows is staggering. Apple had a much easier time forcing developers to port their software and for the rest there's Rosetta 2 (though I still run into the occasional issue at work where an M1 effectively prevents me from getting work done easily).

 

Let's hope Microsoft can pull off something similar. I suppose AMD/Intel would have the advantage of also holding x86 licenses, which might allow them to integrate some form of hardware accelerated emulation. If old software continues to run and new software is also made available for ARM natively, that would certainly be something.

[thevictor390]

45 minutes ago, Eigenvektor said:

The Snapdragon X Elite sounds very promising, at least on paper. But I think the real obstacle to adoption of Windows for ARM is software support, not just raw CPU performance or power efficiency.

 

Unless the software people need/want works on an ARM based machines, I'm not sure it'll get much broader uptake than it has right now. Plus the amount of legacy software on Windows is staggering. Apple had a much easier time forcing developers to port their software and for the rest there's Rosetta 2 (though I still run into the occasional issue at work where an M1 effectively prevents me from getting work done easily).

 

Let's hope Microsoft can pull off something similar. I suppose AMD/Intel would have the advantage of also holding x86 licenses, which might allow them to integrate some form of hardware accelerated emulation. If old software continues to run and new software is also made available for ARM natively, that would certainly be something.

The x86 emulator in Windows has gotten a lot better, but you will still struggle with apps that need lower level access (my most recent fight was a corporate VPN app that couldnt not install its virtual network driver on ARM).

[Monkey Dust]

1 hour ago, Eigenvektor said:

 But I think the real obstacle to adoption of Windows for ARM is software support,

I'm not sure if it will be such a massive issue for many users. A lot of home users only use the browser, and maybe Zoom. Even in offices, some roles only require a browser and MS Office. As long as MS don't screw this up (place your bets ladies and gentlemen) it could meet the needs of most PC buyers. From a PC gaming perspective, I doubt we'll get native ARM games until the consoles go ARM.

 

I wonder if Intel & AMD could pull off hybrid ARM/x64 chips? I guess it would be more if Windows could be made to handle it.

[Zodiark1593]

1 hour ago, Eigenvektor said:

The Snapdragon X Elite sounds very promising, at least on paper. But I think the real obstacle to adoption of Windows for ARM is software support, not just raw CPU performance or power efficiency.

 

Unless the software people need/want works on an ARM based machines, I'm not sure it'll get much broader uptake than it has right now. Plus the amount of legacy software on Windows is staggering. Apple had a much easier time forcing developers to port their software and for the rest there's Rosetta 2 (though I still run into the occasional issue at work where an M1 effectively prevents me from getting work done easily).

 

Let's hope Microsoft can pull off something similar. I suppose AMD/Intel would have the advantage of also holding x86 licenses, which might allow them to integrate some form of hardware accelerated emulation. If old software continues to run and new software is also made available for ARM natively, that would certainly be something.

Tbh, owing to Apple, a lot of productivity applications, and other commonly used applications, already have ARM executables. And a lot of old equipment that needs some old Windows application, will be served by slower, but accurate, emulation. 
 

The only application I think will be royally screwed over is PC gaming, as there’s is a lot of older stuff that will never be ported to ARM, and games are performance-sensitive, and would require substantial investment to bring emulation to sufficient speed. 

[Mark Kaine]

Can't wait for all the "influencers" videos on youtube, "intel destroys nvidia in gaming!"

 

(you know its gonna happen! )

[hishnash]

Makes me wish I was still at my last employer, who was making high end (10k+ per user per year) industrial software for PC.  And it told them they should start to consider getting ready to add ARM support to the low level frameworks they were using so that they would be ready and I got laughed at by the CTO for making a stupid suggestions...  I expect at some point they are going to need to do a all tools down everyone is smashing your heads against the compiler year with no feature development due to not thinking ahead... yes much of the code was old enough to be full of hand crafted inline assembly that will not just compile out. 

[hishnash]

1 hour ago, Zodiark1593 said:

The only application I think will be royally screwed over is PC gaming, as there’s is a lot of older stuff that will never be ported to ARM, and games are performance-sensitive, and would require substantial investment to bring emulation to sufficient speed. 

So at least from apples approach it is not emulation it is called Lifting this is were you have an LLVM interpreter run on the raw compiled code and `lift` it back up to LLVM byte code then you re-target it to a differnt target compile it down to assembly save it to disk and run it (as a poorly optimised arm executable).

The area that some games have issues with is JITs, since these general x86 code at runtime the static lifting does not convert these before the game starts so it needs to do the concretion just in time before you jump to start running that bit of code... this creates stalls.  Games that use LUA etc have this issue (like games from Bethesda) in theory if someone wanted to put in the effort one could write a few patches that swap out the JIT layer within these games so that it writes out ARM64 directly removing the impact, someone like valve would be well placed ot do this if they wanted to (such as use ARM in a future steam deck).  For them they could even do this server side like they pre-compile VK shaders as they know the target HW before you download the game. 

[hishnash]

1 hour ago, Monkey Dust said:

I wonder if Intel & AMD could pull off hybrid ARM/x64 chips? I guess it would be more if Windows could be made to handle it.

Hybrid chips are difficult for a few reasons;

1) memory access ordering is different between standard ARM and x86 
2) you would need a kernel that is mutli arc hybrid... no os kernels exist that can do this today,  macOS is the most dynamic and all it can do is managed switching from 4kb to 16kb page size dynamically (linux cant do it and the work needed would be massive so they have opted for just doing 16kb only) 

 

The only real use case of a hybrid SOC would be if you in effect run a seperate OS on each arc and then use some remote shell system to merge them for the user (a bit like you can have a hybrid VM were some of the applications your using run within teh VM but they show up as if they are standard apps and you can move them around  along with other native apps). 

[captain_to_fire]

1 hour ago, Mark Kaine said:

"intel destroys nvidia in gaming!"

Clickbait cunts such as Max Tech lol

 

 

[Forbidden Wafer]

6 hours ago, hishnash said:

Hybrid chips are difficult for a few reasons;

1) memory access ordering is different between standard ARM and x86 
2) you would need a kernel that is mutli arc hybrid... no os kernels exist that can do this today

Ah, not necessarily. Just look at the PS4 x86 cores for games and compute + arm for background activity and updates. That's actually two kernels, one for each arch, talking to each other and accessing the same exact physical memory. Not that groundbreaking. 

[hishnash]

38 minutes ago, Forbidden Wafer said:

Ah, not necessarily. Just look at the PS4 x86 cores for games and compute + arm for background activity and updates. That's actually two kernels, one for each arch, talking to each other and accessing the same exact physical memory. Not that groundbreaking. 

Yes seperate kernels running software that expects to be running in this situations is very different from a user space application on windows running in x86 alongside another user application running in ARM with neither of them aware of this.  

Having multiple cpus/kernels running talking to the same memory happens all the time, just look at the 12+ little seperate cpus each with thier own kernels that run within the M1 SOC for each of the little bits of it, they all talk to eachtoher over iMessage box and all access the system memory through the same MMU this works without issue but it only works since the software running on these expects to be in this situation, these are no running user-space applications.  

This is very differnt to having a single hybrid kernel that spans both systems and runs user-space code on both. 

[Eigenvektor]

9 hours ago, Monkey Dust said:

I'm not sure if it will be such a massive issue for many users. A lot of home users only use the browser, and maybe Zoom. Even in offices, some roles only require a browser and MS Office. As long as MS don't screw this up (place your bets ladies and gentlemen) it could meet the needs of most PC buyers. From a PC gaming perspective, I doubt we'll get native ARM games until the consoles go ARM.

True. Though I imagine most would only consider a switch if it resulted in considerable cost reduction. Or until they discover the one in-house piece of software that doesn't run properly and is too costly to port (and/or the developer left or died a decade ago…).

 

9 hours ago, Zodiark1593 said:

Tbh, owing to Apple, a lot of productivity applications, and other commonly used applications, already have ARM executables. And a lot of old equipment that needs some old Windows application, will be served by slower, but accurate, emulation.

An app having been ported to Apple Silicon doesn't mean there's no additional effort to port it to Windows for ARM. They are still incompatible platforms, just as they were in the x86 days of Apple.

 

I think in some ways its comparable to Windows vs Linux. Linux could fulfill the needs of those that just need a web browser and even Office (online). But why would you switch if the experience is—at best—the same, rather than substantially better? And you know there's some stuff that doesn't run properly or requires tweaking. Let's see if Windows for ARM can overcome this inertia.

 

My gut tells me you need to get home users on board as quickly as possible. And that means you need to be able to do everything they want, including gaming. For people who can only afford to buy one system, that system needs to be able to do everything. And so far that's only the case with Windows on x86.

[hishnash]

1 hour ago, Eigenvektor said:

My gut tells me you need to get home users on board as quickly as possible. And that means you need to be able to do everything they want, including gaming.

For the avg home user if MS wanted to there is a very easy move... they can just but OEM window license restrictions in place... Windows 12 OEM (consumer licenses) only ship for windows on ARM... if they sign posted this 12 months out and did a little bit extra work on the translation tooling the have for x86 applications, along with get arm cpus with the needed memory modes that make this eaiser (like apple) I don't see them having ot much of an issue.   

Modern games would just support the platform, MS have been doing some interesting work with respect to helping devs that depend on some compiled third party x86 libs by providing some solutions to let the devs ship semi hybrid applications (eg your not always required to have source code or pre-build arm versions of every dependancy to still be able to ship an arm build of what you have an windows translation tool applies to the rest,.. at compile time).   

The issue at the moment is there is no pressure, the games might already compile for arm without issue (most engines do as mobile is rather important for engine devs) but game studios just have not evened bothered to try.

[porina]

I've seen the rumbles for a while but the question at a high level will be, what problem does the move to Arm from x86 solve? Especially if there will be the friction of Windows software compatibility to throw into the mix.

 

Business reasons aside, the three points of the triangle are performance, price, power. Moving towards one area might compromise another.

 

We can partially look at Apple's move. The business side is they wanted more control over the silicon. Their technical claims have leant towards performance and power, and I'd say more towards power. Lower power consumption, with decent performance in chosen use cases. This is partially contributed to from Apple generally being closer to leading edge process nodes than x86 has. Even on similar or same process, Arm for Android hasn't caught up to Apple.

 

Can Windows on Arm reproduce that? I've not looked carefully at existing offerings, but the impression I get is they're lower power, but also lower performance. If the main goal is to provide power efficiency at reasonable performance, the question is do you even have to move to Arm to do that? Can x86 scale down that low if that was the primary design goal? x86 feels like it has traditionally been a performance first design. It needs to go further than E-cores.

[hishnash]

2 hours ago, porina said:

Can x86 scale down that low if that was the primary design goal? x86 feels like it has traditionally been a performance first design. It needs to go further than E-cores.

People look at power draw while under heavy load way to much, for most users 99% of the time there cpus are not under heavy load, this is were the realm magic of ARM chips (including current gen Android and Window ARM) chips shine.  When these chips are under low lower they draw almost 0 power.. less than 1W including memory subsystem, gpu, display engine, IO etc... there is nothing in the x86 space that comes close to this power power state... when im talking low power im not saying standby, I mean when a user is browsing the web, whatchign YouTube etc while the user is using the laptop just not running CB.  Most tasks like web browsing might need some CPU power but it is very short livid most of the time the system almost completely powers down the entier chip.   The fact that the machies are useful while the SOC complex draws less than 1W is what consumer sneed. 

[porina]

1 hour ago, hishnash said:

People look at power draw while under heavy load way to much, for most users 99% of the time there cpus are not under heavy load, this is were the realm magic of ARM chips (including current gen Android and Window ARM) chips shine.  When these chips are under low lower they draw almost 0 power.. less than 1W including memory subsystem, gpu, display engine, IO etc... there is nothing in the x86 space that comes close to this power power state... when im talking low power im not saying standby, I mean when a user is browsing the web, whatchign YouTube etc while the user is using the laptop just not running CB.  Most tasks like web browsing might need some CPU power but it is very short livid most of the time the system almost completely powers down the entier chip.   The fact that the machies are useful while the SOC complex draws less than 1W is what consumer sneed. 

That is a good point. Presuming Arm offerings will be targeted at smaller form factors: thin and light or ultrabooks, maybe even a tablet?

 

It will be interesting to see what Meteor Lake does, as it seems to be preparing in that direction. From memory the SoC tile has two E-cores and media processing, so you don't need to fire up the core tile for light tasks or near idle operation.

[RejZoR]

3 hours ago, porina said:

I've seen the rumbles for a while but the question at a high level will be, what problem does the move to Arm from x86 solve? Especially if there will be the friction of Windows software compatibility to throw into the mix.

 

Business reasons aside, the three points of the triangle are performance, price, power. Moving towards one area might compromise another.

 

We can partially look at Apple's move. The business side is they wanted more control over the silicon. Their technical claims have leant towards performance and power, and I'd say more towards power. Lower power consumption, with decent performance in chosen use cases. This is partially contributed to from Apple generally being closer to leading edge process nodes than x86 has. Even on similar or same process, Arm for Android hasn't caught up to Apple.

 

Can Windows on Arm reproduce that? I've not looked carefully at existing offerings, but the impression I get is they're lower power, but also lower performance. If the main goal is to provide power efficiency at reasonable performance, the question is do you even have to move to Arm to do that? Can x86 scale down that low if that was the primary design goal? x86 feels like it has traditionally been a performance first design. It needs to go further than E-cores.

You can have perfect everything if you control entire stack, top to bottom. Apple does that. They design hardware and software and have own ecosystem around it. What can HP or Dell do for their laptops? They supply hardware from AMD, NVIDIA and Intel, they rely on Microsoft to do software with software ecosystem almost entirely out of everyone's control but Microsoft's who's not the designer of hardware it's running on. And there is this disconnect that's less than optimal. Apple doesn't have those issues at all. Downside is that you have to keep on track of everything to make it work. If you have issues, you'll have issues top to bottom in the end, but if you don't, you can have way better results. Example is for example Intel and their chip foundry. When they struggled with fab process, their chips suffered and they couldn't do anything about it, meanwhile AMD relied on TSMC and it just worked. Issue here is, if TSMC had issues, AMD would be at their mercy. So it's not always a clear cut thing. But in general, if it work, it works better.

[leadeater]

1 hour ago, hishnash said:

People look at power draw while under heavy load way to much, for most users 99% of the time there cpus are not under heavy load, this is were the realm magic of ARM chips (including current gen Android and Window ARM) chips shine.  When these chips are under low lower they draw almost 0 power.. less than 1W including memory subsystem, gpu, display engine, IO etc... there is nothing in the x86 space that comes close to this power power state... when im talking low power im not saying standby, I mean when a user is browsing the web, whatchign YouTube etc while the user is using the laptop just not running CB.  Most tasks like web browsing might need some CPU power but it is very short livid most of the time the system almost completely powers down the entier chip.   The fact that the machies are useful while the SOC complex draws less than 1W is what consumer sneed. 

I don't think it's the SoC between Apple and x86 that is the battery life problem for laptops, not for that type of usage. The bigger problem is everything else in the laptop, the stuff not in the SoC, the choice and calibration of screen for power, wifi calibration for power etc. All that other stuff adds up and it's where I think Apple laptops have the biggest advantage in low power rather than just looking at SoC power. You can run a test like playing YouTube and both SoC's will be drawing almost no power as both are offloading to very power efficient hardware decoders and Zen3 cores power gate park at less than 0.1W per core and the mobile parts aren't chiplet except for the 7045 series so the minimum power tradeoff for a IOD+CCD doesn't really exist on laptops.

 

4k YouTube playback:

Asus ZenBook S 13: 7.7W

MacBook Pro 13: 4.5W

 

BTW PCMark 10 battery life test "Applications" the above Asus laptop power draw is 5.5W average (12.18 hours)

 

And the above difference is not all SoC. The other problem in x86 land is the same CPU in similar laptops themselves have different power characteristics and the devices can have quite drastic difference in runtime tests. Apple's biggest benefit is consistency, there is only 1 of XYZ product (our one), which I think matters a lot for the final products.

 

P.S Add in a dGPU and you can kiss power efficiency goodbye.

[wanderingfool2]

7 hours ago, Eigenvektor said:

True. Though I imagine most would only consider a switch if it resulted in considerable cost reduction. Or until they discover the one in-house piece of software that doesn't run properly and is too costly to port (and/or the developer left or died a decade ago…).

One other thing that it could end up doing though is better performance for certain tasks as well though.  So that might be a way into the market as well.

 

After all, the x86-64 is such a bloated set of instructions there is a ton of wasted space on the CPU die for essentially instructions that no one is realistically is going to use (except in a few edge case programs).  I think that's where MS will be able to just to x86 emulation, I'd imagine most wouldn't be too hard...just those programs that use that assembly would take massive performance hits.

 

Honestly though, the biggest win I could see for ARM, aside from making actual low powered CPUs that perform well is the ability to make one that's quite a bit more powerful for everyday kind of tasks (with exceptions for things like emulations which do require some of those x86 bloated instructions)

[porina]

13 minutes ago, wanderingfool2 said:

After all, the x86-64 is such a bloated set of instructions there is a ton of wasted space on the CPU die for essentially instructions that no one is realistically is going to use (except in a few edge case programs).  I think that's where MS will be able to just to x86 emulation, I'd imagine most wouldn't be too hard...just those programs that use that assembly would take massive performance hits.

That's already happening in x86 CPUs. The instructions and execution units are decoupled. Instructions are broken down into micro-ops, and fed to appropriate execution units for the desired result. Execution resource that isn't well used may be removed and are in effect emulated for those instructions that use it, with some performance penalty. Arm translating x86 has to do pretty much the same thing. It isn't a saving.

[leadeater]

1 hour ago, wanderingfool2 said:

After all, the x86-64 is such a bloated set of instructions there is a ton of wasted space on the CPU die for essentially instructions that no one is realistically is going to use (except in a few edge case programs). 

Eh not really, can't remember which person in the industry commented on this but they said all the redundant not used instructions take up next to nothing which is also why they don't get taken out. If they really did they'd have a direct attributable cost to Intel/AMD in silicon fabrication and would get axed real fast.

 

 

uCode ROM is where the instruction set are and that die area is the entire microcode for everything, it's very doubtful taking away 25% of the instructions that exist, used or not, would get close to 25% area reduction.

 

If I can find the source quote on this I'll update, but suffice to say the above does show that even zoomed in to a single core perspective it's not a lot.

 

The decode also has to handle any instructions too so there is that part too but I don't see much space savings happening there either. 

 

Edit:

Intel's x86-S proposal is to entirely drop 16bit and 32bit support which is why it would actually save some die space and also everything would have to be 64bit only basically forcing cleanup of many things in the process of having to because your old "shit" won't work anymore heh. Note 32bit user applications would still work.

 

https://www.hwcooling.net/en/x86-s-intel-wants-to-drop-legacy-compatibility-from-processors/

 

https://www.intel.com/content/www/us/en/developer/articles/technical/envisioning-future-simplified-architecture.html

[porina]

30 minutes ago, leadeater said:

Eh not really, can't remember which person in the industry commented on this but they said all the redundant not used instructions take up next to nothing which is also why they don't get taken out. If they really did they'd have a direct attributable cost to Intel/AMD in silicon fabrication and would get axed real fast.

Jim Keller has said words to the effect the maturity of a specific ISA doesn't make much of a difference, in multiple interviews. I don't know if he might have said what you said elsewhere.

[leadeater]

1 minute ago, porina said:

Jim Keller has said words to the effect the maturity of a specific ISA doesn't make much of a difference, in multiple interviews. I don't know if he might have said what you said elsewhere.

Was more than likely him yea.

[dizmo]

I'm more interested to see what Intels response is. They must have some kind of plan, and if they don't it shows their inflexibility as a company which usually leads to failure. 

 

Competition is great and all but I highly doubt we'll see lower pricing. We'll just have more choice.