News Nvidia and AMD to Develop Arm CPUs for Client PCs: Report

[setx]

AMD likes ARM because it has inherent efficiency advantages over x86. Because of that, nobody can compete with Apple on perf/W and ARM represents a major competitive threat in the cloud.

ARM "efficiency advantages" are overrated. Software optimization is far more impactful and so far x86 software is better optimized.

Apple's performance has nothing to do with ARM in particular. It's from tight integration/mutual optimization of hardware/OS/software.

And their "competitive threat in the cloud" remains only a threat for many years already.

In the end, regardless of any technical advantages, ARM's licensing behavior kills the interest of any sane major company to start developing their hardware. And who already tied to ARM can't just exit as switching ISA takes several years and RISC-V isn't quite ready yet.

 

[bit_user]

ARM "efficiency advantages" are overrated. Software optimization is far more impactful and so far x86 software is better optimized.

Apple's performance has nothing to do with ARM in particular. It's from tight integration/mutual optimization of hardware/OS/software.

Intel seems to disagree. With APX, they copied 3 of AArch64's 5 main advantages and claim it gain them 10% more performance. That's nontrivial.

And their "competitive threat in the cloud" remains only a threat for many years already.

Well, Amazon's Graviton is already on its 3rd gen. Probably gen 4 will be announced any day now. Also, Nvidia is shipping their ARM-based "Grace" CPUs, which should replace x86 processors in a lot of AI-optimized machines.

In the end, regardless of any technical advantages, ARM's licensing behavior kills the interest of any sane major company to start developing their hardware.

Ah, yes. That's the main issue facing someone in the chip business, like Qualcomm, AMD, or Nvidia. However, AMD and Nvidia already have architectural licenses. ARM can't change those until they come up for renewal. So, the one potentially left out in the cold is Qualcomm.

And who already tied to ARM can't just exit as switching ISA takes several years and RISC-V isn't quite ready yet.

Yeah, but these ARM-based processors aren't due until 2025. By then, RISC-V in the cloud will be much closer, if not already a reality.

 

[bit_user]

I've never heard of AMD bringing ARM to AM2/AM3/AM4 which became retail products or even available on eBay.

Note: Pretty much all Opterons do not fit into mainstream AMx sockets. I think I was pretty clear that I meant mainstream sockets.

As was mentioned above, there was Skybridge, but that wasn't released that I know of.

Yeah, it turns out I was indeed thinking of Skybridge. I didn't realize that was something different than the A1100 which eventually came to market.

I never meant to suggest they shipped ARM CPUs that plugged into a client (AMx) socket.

 

[RedBear87]

I know Samsung also sells Chromebooks, but not sure if they use their own SoC or others'.

Samsung's Chromebooks are all Intel based, actually there were rumours about some kind of derivative of the RDNA2 equipped Exynos 2200 coming to laptops, but so far nothing has materialised. There were also more recent rumours about Mediatek expanding to the mobile and laptop sectors its partnership with Nvidia, currently limited to the automotive sector. To be honest AMD directly entering this market looks especially puzzling to me, I can't help wondering whether this Reuters report is omitting/missing that they're partnering with another company.

 

[TerryLaze]

Yeah, these will be desktop PCs in the same way that the intel CEO that claimed that meteor would be coming to desktop PCs....

For nvidia it's super clear that it's better to be able to make a SoC completely on their own instead of having to pay anybody else for the CPU part.

For AMD I don't see this making much sense, they have products with good margins, they even stopped making low end x86 CPUs, why would they start making ARM CPUs with even lower margins than the desktop CPUs they already stopped making?!

Also intel will be happy to FAB these CPUs for nvidia or even AMD if they choose to make them.

Intel Foundry and Arm Announce Multigeneration Collaboration on...

Collaboration to bring chip designers a powerful combination of Arm core and Intel angstrom-era process technology advancements.

www.intel.com

 

[pug_s]

IMO, Microsoft's greatest strength of trying to maintain combability is its greatest weakness in terms to port to another ISA. The problem is not the Windows kernel and OS itself but the runtime and an alphabet soup of plugins is the problem. Think of the Visual C++ (year) runnable, visual basic (year), .net, office 365 runtime, java and etc... This additional layer makes it easier for the developers to make software easier to code, but it is just less portable. You don't see Google's Chrome, and Apple IOS not supporting that stuff and forcing developers to update their code which simply does not happen to Microsoft.

 

[RichardtST]

If the ARM instruction set wasn't so annoying I'd root for them. I mean, come on. Who really needs a barrel shifter anyway? Maybe if they hid it better under a more sensible assembly language... I still think Intel and AMD would be far better off making a completely stripped x86 instruction set that only had the instructions necessary for the 90%.

 

[Order 66]

until ARM becomes completely compatible with windows applications and games (I know they are fundamentally different so maybe it will never happen) I will not be using anything with arm on the desktop PC side of things.

 

[JamesJones44]

When the software they want doesn't work on Windows on ARM, they'll abandon the platform for a x86 based Windows.

That's is such a tiny sample of the overall product. These days any software built in the last 10 years is in a lot of cases a simple recompile.

I don't know that many people or companies that cling to software that is 15+ years old. Yes, they exist, but it's not 80% by a long shot.

 

[JamesJones44]

And even Apple's Rosetta leaves many users to prefer the original hardware when running x86 Software.

I don't know about that. It's fairly rare to find software on Mac that still requires Rosetta these days. Yes, there is a performance hit while Rosetta is there, but most software is native apple silicon these days. However, even with the average 30% Rosetta 2 hit, it still performs better than a lot of the x86 direct equivalent CPUs out there. It's hard to say of other ARM CPUs will perform as well, current Qualcomm ones currently don't, but if they can ever release Nuvia based processor that could change.

The 2020 Mac Mini Unleashed: Putting Apple Silicon M1 To The Test

www.anandtech.com

The bigger thing I would point out is that it has only be 3 years and most of the macOS echo system has been converted (most of it was converted by year 2). The Windows eco system will likely take twice as long, but I still expect it would be quick given the tools that are available to developers now.

 

[Kamen Rider Blade]

That's is such a tiny sample of the overall product. These days any software built in the last 10 years is in a lot of cases a simple recompile.

I don't know that many people or companies that cling to software that is 15+ years old. Yes, they exist, but it's not 80% by a long shot.

Tell that to the average customer who isn't well informed on what's available on the market or has a key piece of software for their workflow that they can't or won't replace for whatever reason.

The bigger thing I would point out is that it has only be 3 years and most of the macOS echo system has been converted (most of it was converted by year 2). The Windows eco system will likely take twice as long, but I still expect it would be quick given the tools that are available to developers now.

And if the Windows Customer Base refuses to buy or use ARM based Windows?

Then what?

More Windows RT based LapTops sitting on store shelves, depreciating?

 

[setx]

Intel seems to disagree. With APX, they copied 3 of AArch64's 5 main advantages and claim it gain them 10% more performance. That's nontrivial.

Disagree that software optimization is the most important? Definitely not. Those new instructions won't be magically used by old software. Pretty much the only definite advantage of ARM ISA is their simpler instruction decoding logic and APX, sadly, has nothing to do with that. Sure, more registers is nice but how much that affects performance in modern superscalars is hard to tell. 2op vs 3op arithmetic issue is already kind of solved with almost free mov.

which should replace x86 processors in a lot of AI-optimized machines.

Pretty much anything can be CPU in "AI-optimized machines", but looking at Google it's going to be RISC-V, not ARM.

 

[jayjr1105]

Where does RISCV stand in this picture? I feel like it's going to replace ARM at some point being royalty/license free.

 

[jackt]

arm or riscv ? well riscv is amd rival... arm too...

 

[JamesJones44]

Tell that to the average customer who isn't well informed on what's available on the market or has a key piece of software for their workflow that they can't or won't replace for whatever reason.

Like what? Please give an example of something you think the average business or user will try to run that won't.

Adobe already did this transition Mac, Microsoft has done it for Mac, AutoDesk has done it Mac, etc. Popular developer tools like Electron (threw up a bit in my mouth on that, but people still use it), .NET, Java, Rust, etc already did this transition with Mac. As you can see, several big players in the software space did this already for Mac. It's not hard to see them being able to do it for Windows in less time since they now have experience and the tools to make it happen quickly.

This is also ignoring translation/emulation. If the CPU is 30% faster the 30% reduction in speed isn't going to be that big of a deal.

 

[JamesJones44]

Where does RISCV stand in this picture? I feel like it's going to replace ARM at some point being royalty/license free.

I don't know if we are there yet on RISC-V but fast forward 5 years and I wouldn't be surprised to see someone attempt it. Qualcomm has talked about using it for Android Wear processors. That's a fair step considering that has also been traditionally dominated by ARM.

 

[Kamen Rider Blade]

Like what? Please give an example of something you think the average business or user will try to run that won't.

Adobe already did this transition Mac, Microsoft has done it for Mac, AutoDesk has done it Mac, etc. Popular developer tools like Electron (threw up a bit in my mouth on that, but people still use it), .NET, Java, Rust, etc already did this transition with Mac. As you can see, several big players in the software space did this already for Mac. It's not hard to see them being able to do it for Windows in less time since they now have experience and the tools to make it happen quickly.

This is also ignoring translation/emulation. If the CPU is 30% faster the 30% reduction in speed isn't going to be that big of a deal.

It's usually smaller software that are proprietary where the devs don't support it anymore and they have expensive machinery that is dependent on that software to run.

 

[JamesJones44]

It's usually smaller software that are proprietary where the devs don't support it anymore and they have expensive machinery that is dependent on that software to run.

These same types of concerns were echoed for 64-bit only operating systems and it turned out to be a tiny fraction of the overall base (like 5 or 6% if I recall). If that software is new enough to run on 64 bit only then I don't think there will be any issues.

Even if it is, Microsoft isn't going to get overly concerted with 10% or less of the market share having issues. Those users can stick with x86 or use emulation/translation.

 

[NinoPino]

Apple's Rosetta 2 can emulate at like 90% the speed of natively-compiled.

90% is a bit irrealistic also for a good JIT emulator.

 

[vinay2070]

nVidia getting into this is what excites me. They already have some experience in ARM CPUs. If they can come up with a good M3 like ARM based CPU with 4070 class graphics with integrated high speed memory on a windows Laptop, I would definitely buy it. Probably a few years away, but one can dream.

 

[cambrown]

I see what MS wants from ARM (to threaten Intel), but what is there for AMD? Razor-thin margins? Paying more to ARM for nothing? Zen was developed as dual x86/ARM architecture but ARM part was abandoned later.

They want to compete in the laptop space with Apple, quite possibly. There is another case to be made for AI acceleration on ARM since it's so much less power consumptive than x86 at equivalent implementation scales.

Also, having better perf/watt means greater capacity for multicore designs given fixed power constraints.

Also, your comment about "efficiency advantages" of ARM have no basis in reality. You're proved wrong empirically by Apple's perf/watt as well as in the datacenter by Ampere's ARM based chips vs. AMD/Intel x86 offerings. If it was possible to achieve the power savings purely through software optimization, why did Apple switch to ARM? Why do all mobile processors run on ARM? Just a preposterous take.

 

[cambrown]

Where does RISCV stand in this picture? I feel like it's going to replace ARM at some point being royalty/license free.

RISC-V will not be a player for a while. There will be incremental gains, just as there were with ARM. However, its ecosystem is not mature enough for full adoption. SiFive just laid off a ton of people, and are switching to a customized core-based business model it would seem. Definitely good to watch out for over the next 10 years though, because it very well could be in the position the ARM ISA finds itself now.

 

[bit_user]

Samsung's Chromebooks are all Intel based,

Not always.

• https://arstechnica.com/gadgets/201...-arm-chromebook-gets-by-without-intel-inside/
• https://www.samsung.com/uk/support/model/XE303C12-H01UK/

actually there were rumours about some kind of derivative of the RDNA2 equipped Exynos 2200 coming to laptops, but so far nothing has materialised.

I'm pretty sure their RDNA licensing agreement with AMD prevents them from using it to directly compete with AMD. I've read that, at least. The logical conclusion of that is they couldn't use it in chromebooks, if AMD were also in the chomebook market.

 

[bit_user]

For AMD I don't see this making much sense,

The article gives a good reason. If Microsoft is already decided on using ARM for the next XBox, then maybe that was enough to push AMD into the game. Then, they figured they might just go ahead and also put it in some mini-PCs or something, because "why not?"

If the Windows-on-ARM market fails to materialize, they'll still have XBox as a way to recoup their design costs and can simply cancel the desktop version.

 

[bit_user]

IMO, Microsoft's greatest strength of trying to maintain combability is its greatest weakness in terms to port to another ISA.

Well, Windows-on-ARM has been a thing for several years, at least. Did you know that Windows NT once ran on MIPS, DEC Alpha, PowerPC, IA64, and perhaps others? Some of those are big-endian, even! Plus, Windows Mobile/Phone obviously runs on plenty of ARM SoCs.

Microsoft does know a few things about porting to other ISAs!

Think of the Visual C++ (year) runnable, visual basic (year), .net, office 365 runtime, java and etc... This additional layer makes it easier for the developers to make software easier to code, but it is just less portable.

Why is it less portable? Java was designed to be super-portable, by using an intermediate bytecode representation, instead of natively compiling programs for a single ISA. With C# and .Net, Microsoft copied that idea. So, I'll bet a lot of Windows apps just run on ARM with no porting or recompilation necessary.

You don't see Google's Chrome, and Apple IOS not supporting that stuff and forcing developers to update their code which simply does not happen to Microsoft.

Uh... you lost me. Can you rephrase that, please?