Wednesday, November 17th 2021

Qualcomm Says PC Transition to Arm-based Processors is Certain, to Launch High-Performance SoCs in 2023

Nov 17th, 2021 02:40 Discuss (53 Comments)

Qualcomm has been in the game of creating SoCs for the PC market with the company's Snapdragon lineup. These processors mainly were beefed-up versions of their mobile designs and were based on the Arm instruction set architecture (ISA). Microsoft has backed this effort by creation Windows-on-Arm (WoA) project that enables the Windows OS to operate on Arm processors. However, up until now, Qualcomm's designs were not very powerful as they represented a relatively moderate approach to the problem and almost made no sense of purchase compared to the standard laptops equipped with x86 processors from AMD and Intel. This is about to change.

According to the news from Investor Day yesterday, Qualcomm is preparing high-performance Arm SoCs for the PC market. The company has recently acquired Nuvia Inc., a startup focused on creating novel IPs based on Arm ISA. And this is what Qualcomm will use in building its next-generation PC processors. As the company plans, in August of 2022, it should start sampling OEM partners with these new chips, and we will be seeing them in consumers' hands in early 2023. If everything goes as planned, this should represent direct competition to AMD, Intel, and now Apple in the high-end SoC market. After PCs, the company plans to tackle datacenter, mobile, and automotive market.

Sources: via Tom's Hardware, Andreas Schilling (Image)

Add your own comment

53 Comments on Qualcomm Says PC Transition to Arm-based Processors is Certain, to Launch High-Performance SoCs in 2023

Vayra86

GL with that. Microsoft says no. And not for lack of trying.

Flanker

They've been going on about this for how long now?

Valantar

I voted "certainly", but on the condition that it actually performs at least as well as X86 and delivers some sort of benefit - power efficiency being the most obvious one. Apple is demonstrating that it can be done, but they are doing it with gargantuan R&D budgets, full vertical integration, semi-exotic packaging technologies, giant dice, and more. It still remains to be seen if anyone else is able to get even remotely close. QC and ARM have so far not dared to make a core even close in size (both physical and architecturally), and it still remains to be seen if they will do so. If not, the answer is a clear no.

Vya Domus

No.

And even if it were true Mircrosoft sure doesn't thinks so, the x86 emulation on the ARM branch of Windows is sub-par, you can only run 32-bit applications and even then it's full of issues. Plus, it doesn't look like it will improve anytime soon and without that ARM has no future on PCs.

Flanker

Valantarfull vertical integration

This is what sets Apple aside from everyone else imo

Valantar

FlankerThis is what sets Apple aside from everyone else imo

That, and their heaping pile of cash. One could argue the latter is how they achieved the former.

Vya DomusNo.

And even if it were true Mircrosoft sure doesn't thinks so, the x86 emulation on the ARM branch of Windows is sub-par, you can only run 32-bit applications and even then it's full of issues. Plus, it doesn't look like it will improve anytime soon and without that ARM has no future on PCs

It's clear that things need drastic improvements for this to be viable. Both more apps compiled natively for ARM (which should be helped by MS's less draconian Store requirements these days) and better emulation. The Apple M1 reportedly has significant hardware-level tweaks to improve its ability to emulate x86.

silentbogo

I'm wondering what's the delay?
They've been teasing Windows on ARM demos with MS for half-a-decade now, yet still ARM can only be found in only in few chromebooks and even fewer niche laptops and hybrids...
I'm not sure what's happening with this whole thing, but it all kinda died down. QC stopped pushing X1, MS moved x64 emulation to Win11, Snapdragon ultrabooks went into hiding, and now - delays.

Vya DomusPlus, it doesn't look like it will improve anytime soon and without that ARM has no future on PCs.

I guess someone missed the whole Apple M1 ordeal. It can work and it works, and given a couple of years devs will get a hang of creating native ARM apps for desktops.
Heck, even old MS demos were quite impressive (though 32-bit only), and that's not including the fact that quite a few native Windows on ARM ports of popular software have been released since then.

bug

The poll is not well thought out. I, for one, couldn't care less about the underlying uarch. I'm mostly on Linux and Linux supports (almost) everything under the sun. But people on Windows aren't that lucky. There are a lot of old apps out there, some of them abandoned, but they still work. Unless Microsoft bothers with a translation layer, moving will be difficult.

londiste

Vayra86GL with that. Microsoft says no. And not for lack of trying.

The interesting bit is that Microsoft actually has tried Windows on ARM on a couple different occasions, focusing on cheap/mobile that is the strong point of ARM SoCs. Despite limitations the attempts seem to have been earnest. Currently the biggest pain seems to be x86 and its vast software library. As Microsoft does not have full vertical control like Apple nor seems to want that, x86 emulation exists but getting a performant one is tricky enough to achieve.

ValantarI voted "certainly", but on the condition that it actually performs at least as well as X86 and delivers some sort of benefit - power efficiency being the most obvious one. Apple is demonstrating that it can be done, but they are doing it with gargantuan R&D budgets, full vertical integration, semi-exotic packaging technologies, giant dice, and more. It still remains to be seen if anyone else is able to get even remotely close. QC and ARM have so far not dared to make a core even close in size (both physical and architecturally), and it still remains to be seen if they will do so. If not, the answer is a clear no.

M1 power efficiency is a bit overstated. It is very efficient but M1 is a whole node ahead at 5nm compared to AMD/Intel, is designed for and runs at a low voltage/power and is quite large in terms of transistor/area budget. It is ahead but not by too much.

#10

Mussels

Freshwater Moderator

I truly believe Intel will start a trend with the hybrid processor designs, and this is where we will end up - mixing Arm and x86 in the same systems with windows 11 and its support for hybrid, mixed architectures

Theres no reason Intels E-core concept cant become mainstream, with some highly optimised efficient cores taking care of the OS and lightweight tasks, with performance cores smashing out the big stuff as needed.

#11

Valantar

silentbogoI'm wondering what's the delay?
They've been teasing Windows on ARM demos with MS for half-a-decade now, yet still ARM can only be found in only in few chromebooks and even fewer niche laptops and hybrids...
I'm not sure what's happening with this whole thing, but it all kinda died down. QC stopped pushing X1, MS moved x64 emulation to Win11, Snapdragon ultrabooks went into hiding, and now - delays.

The X1 core is in every high end ARM-based design outside of Apple though - Qualcomm, Samsung, even MTK I think? The thing is, it's not that fast. It's faster than A78, but not by a ton. It's by no means an M1 Firestorm core. Heck, the low power Icestorm cores aren't that far behind (and absolutely trounce the A53). They need to go much, much bigger in their designs to compete. Anandtech gets into this in their M1 deep dive articles, which show why the M1 is just that much faster. It's not magic, but mostly down to them being willing and able to make a much, much larger design with huge caches and tons of execution ports - and doing it well. ARM/QC needs to get out of the realm of tiny cores and dare to spend some area (=money) to compete.

#12

cyneater

Yarn...

Need a laugh emoji

If arm wasn't SOC maybe ...
Eg upgradable processors motherboards etc...

PC is about all the parts and bits and pieces you can do.
Buying a prebuilt SOC is pretty bland and boring.

Also big risc of the 90's .... Yet we are still using X86

#13

Vayra86

MusselsI truly believe Intel will start a trend with the hybrid processor designs, and this is where we will end up - mixing Arm and x86 in the same systems with windows 11 and its support for hybrid, mixed architectures

londisteThe interesting bit is that Microsoft actually has tried Windows on ARM on a couple different occasions, focusing on cheap/mobile that is the strong point of ARM SoCs. Despite limitations the attempts seem to have been earnest. Currently the biggest pain seems to be x86 and its vast software library. As Microsoft does not have full vertical control like Apple nor seems to want that, x86 emulation exists but getting a performant one is tricky enough to achieve.

M1 power efficiency is a bit overstated. It is very efficient but M1 is a whole node ahead at 5nm compared to AMD/Intel, is designed for and runs at a low voltage/power and is quite large in terms of transistor/area budget. It is ahead but not by too much.

This.

I do think we're going to be looking at such baby steps, once the stuff is out in the wild, that nobody will care.

People want stuff that 'just works'. x86 - just works. And it is so expansive in what it offers, ARM 'content' can't hold a candle to it, even with a dozen App Stores feeding it. The vast majority of stuff people do on ARM is also not productivity, or at least 'less productive'. People know ARM as 'smartphone'.

Inevitably, if ARM wants to either emulate or mimic x86, its going to be the same thing, equally fat and bloated and the advantage will still be architectural rather than choosing 'this or that'. Yay it can work with cloud, but that also means nothing is truly ARM, it just works on it as it could elsewhere.

ARM gets there when we feel it has feature AND content parity. Anyone saying it will prior to that, is dreaming.

#14

Valantar

londisteM1 power efficiency is a bit overstated. It is very efficient but M1 is a whole node ahead at 5nm compared to AMD/Intel, is designed for and runs at a low voltage/power and is quite large in terms of transistor/area budget. It is ahead but not by too much.

Well, one could argue that it being huge is exactly why it's ahead, rather than this somehow being a point against its efficiency. That makes it expensive, and its performance per area is likely much less impressive, but again, that's what tons of money and vertical integration gets you. Now, don't get me wrong, the M1 Pro/Max in the MBP can hit near 100W power draw under heavy CPU+GPU loads, so it can definitely chug power if it wants to - but it also more or less matches 20W Zen3 cores and 50+W Golden Cove cores in ST performance around 11W/core. Which also lets it scale impressively for MT. It's essentially the same advantage that AMD had played on with Zen vs Intel - not quite there in ST, but superior efficiency lets them scale MT without dropping clocks too badly.

#15

Mussels

Freshwater Moderator

cyneaterYarn...

Need a laugh emoji

If arm wasn't SOC maybe ...
Eg upgradable processors motherboards etc...

PC is about all the parts and bits and pieces you can do.
Buying a prebuilt SOC is pretty bland and boring.

Also big risc of the 90's .... Yet we are still using X86

It's no different to how chipsets on a motherboard are now.

Hell, what if future chipsets had ARM E-cores inside them (So a higher end mobo had faster E-cores) and would boot and run the OS just fine like that, and we slapped in performance cores just like we do GPU's.

#16

Vayra86

ValantarWell, one could argue that it being huge is exactly why it's ahead, rather than this somehow being a point against its efficiency. That makes it expensive, and its performance per area is likely much less impressive, but again, that's what tons of money and vertical integration gets you. Now, don't get me wrong, the M1 Pro/Max in the MBP can hit near 100W power draw under heavy CPU+GPU loads, so it can definitely chug power if it wants to - but it also more or less matches 20W Zen3 cores and 50+W Golden Cove cores in ST performance around 11W/core. Which also lets it scale impressively for MT. It's essentially the same advantage that AMD had played on with Zen vs Intel - not quite there in ST, but superior efficiency lets them scale MT without dropping clocks too badly.

M1 works because it works in a vacuum, that's really the gist.

In the real world not designed by Apple, that won't fly.

#17

Valantar

MusselsI truly believe Intel will start a trend with the hybrid processor designs, and this is where we will end up - mixing Arm and x86 in the same systems with windows 11 and its support for hybrid, mixed architectures

Can that even work? Won't that necessitate every application to be installed twice, essentially, with one version compiled for ARM and one for X86? Unless you're talking about emulation on the ARM cores, which begs the question if a low power x86 core wouldn't run far more efficiently due to less overhead.

The E cores have a big future though. I was very skeptical beforehand, but their performance is very convincing, as is their efficiency. I'm looking forward to seeing a 2P+4E or even 2P+8E mobile ADL design. That would be really interesting.

Vayra86M1 works because it works in a vacuum, that's really the gist.

In the real world not designed by Apple, that won't fly.

I get what you're trying to say, but MacOS is very much the real world. It's a semi-controlled environment, but one that runs a wide variety of real software, and it works there too.

Heck, the fact that you can run Windows in Parallels on an M1 and have it kind of work decently is downright mind-boggling.

#18

Vya Domus

silentbogoI guess someone missed the whole Apple M1 ordeal.

I think you missed it. Apple provided comprehensive emulation support when they moved to ARM, Microsoft didn't. It doesn't matter how good the hardware is, if you can't run every application without issues no one is going to want to switch over.

#19

Vayra86

ValantarI get what you're trying to say, but MacOS is very much the real world. It's a semi-controlled environment, but one that runs a wide variety of real software, and it works there too.

Heck, the fact that you can run Windows in Parallels on an M1 and have it kind of work decently is downright mind-boggling.

Its also a mere 10-12% of the marketplace. Apple's effort is for good reasons, they NEED to connect to matter, especially as content flies between different devices.

So its the real world, from the Apple perspective, where the margins are huge enough to keep fine tuning software to cater to hardware. A somewhat different economic reality. Compared to a wide invasion of ARM in x86, its an entirely different dimension.

#20

Fourstaff

People are using Chromebooks and kitting their iPad with keyboards, so ARM is definitely slowly creeping into the laptop space. I don't think it will replace x86 short of a heroic attempt by Microsoft to create their version of Rosetta, the same way how a lot of legacy systems are still using POWER and older architectures. The number of people using x86 will keep dropping as ARM's software support gets fleshed out.

#21

Valantar

Vayra86Its also a mere 10-12% of the marketplace. Apple's effort is for good reasons, they NEED to connect to matter, especially as content flies between different devices.

So its the real world, from the Apple perspective, where the margins are huge enough to keep fine tuning software to cater to hardware. A somewhat different economic reality. Compared to a wide invasion of ARM in x86, its an entirely different dimension.

Absolutely - that's why I pointed out the systemic advantages (and the accompanying needs/desires that come with these) that Apple has above. Vertical integration and loads of cash lets you do certain things in ways that others can't. But that doesn't take away from the fact that if the rest of the ARM world wants to even remotely keep up, they need to try to follow suit. Obviously they need to keep it economically feasible, but we've seen just how little of an impact a half-measure like X1 has in this reality. The challenge is how to make a huge core design that is still economically feasible for a non-integrated market. IMO a huge portion of this will be to also have an actually fast small core, as the A53 is woefully slow by today's standards, making Apple-like 2+4 (or similar) designs perform very poorly. Thus most ARM vendors need more big cores, which doubly disadvantages them.

#22

Vya Domus

ValantarHeck, the fact that you can run Windows in Parallels on an M1 and have it kind of work decently is downright mind-boggling.

A far as I know only the ARM version of Windows works in Parallels which means that it's technically running natively ?

#23

Valantar

Vya DomusA far as I know only the ARM version of Windows works in Parallels which means that it's technically running natively ?

Hm, it's quite possible that I've missed that part in the short demos I've seen. But won't that still include MS's Win32 emulator, if so?

#24

Vya Domus

ValantarBut won't that still include MS's Win32 emulator, if so?

I guess, I am not really that impressed by the emulation to be honest, there are things that work in favor of x86 emulation on ARM, like the fact that you have more general purpose registers available.

By the way, I still don't understand how Apple got away with x86 emulation, Intel cracked down on every big corporation who wanted to do that in the past, like Nvidia and even Microsoft. You'd think they would do it to the one company where it would actually matter.

#25

londiste

Valantarit also more or less matches 20W Zen3 cores and 50+W Golden Cove cores in ST performance around 11W/core.

Have you seen clock-for-clock (or power-limited) benchmarks with M1 vs Zen3 vs something from Intel somewhere? I am really curious about how that would work out.

Zen3 at 20W should run at 5GHz, Golden Cove at 50+W runs at 5.2-5.3GHz, both are on a quite steep curve or at the end of reasonable curve at that point. At 11W Zen3 should be around 4.2GHz and Golden Cove at around 3.9GHz.

Also relevant - TSMC claims 20% more performance or 40% less power for N5 over N7.

ValantarBut that doesn't take away from the fact that if the rest of the ARM world wants to even remotely keep up, they need to try to follow suit.

Looking at this angle - and only at this angle - Nvidia buying ARM might actually be very beneficial. Nvidia should have both the resources and knowhow to make that happen and software support is their strong side.

ValantarThe challenge is how to make a huge core design that is still economically feasible for a non-integrated market. IMO a huge portion of this will be to also have an actually fast small core, as the A53 is woefully slow by today's standards, making Apple-like 2+4 (or similar) designs perform very poorly. Thus most ARM vendors need more big cores, which doubly disadvantages them.

ARM will now have to contend with Gracemont and soon Zen4c/Zen4e. Outside Apple that picture is not looking too rosy.

Add your own comment

Qualcomm Says PC Transition to Arm-based Processors is Certain, to Launch High-Performance SoCs in 2023

53 Comments on Qualcomm Says PC Transition to Arm-based Processors is Certain, to Launch High-Performance SoCs in 2023

Latest GPU Drivers

New Forum Posts

Popular Reviews

Controversial News Posts

Qualcomm Says PC Transition to Arm-based Processors is Certain, to Launch High-Performance SoCs in 2023

Related News

53 Comments on Qualcomm Says PC Transition to Arm-based Processors is Certain, to Launch High-Performance SoCs in 2023

Latest GPU Drivers

New Forum Posts

Popular Reviews

Controversial News Posts