Qualcomm Nuvia vs Apple Mx

[cmaier]

So its quite possible that a Nuvia chip won’t match the single-threaded performance of M1, just because they haven’t been as focussed on that aspect of performance. Interesting.

Anyway I’d expect the M2 to come in 2022, but likely the M3 not until 2024.

Well I imagine they had to start over when Qualcomm acquired them (or perhaps shortly before). Anyway, no way of knowing. But one hill I’ll die on is that Nuvia should not be assumed to be any sort of miracle for Qualcomm. They never made a chip. Their marketing numbers aren’t very meaningful unless you believe that it’s easy to produce a physical design that lives up to your high level concept. The fact that Qualcomm paid a lot of money for them does not mean they were worth the money. And talking up a chip that won’t be on the market for a year is probably a bad idea.

They may make a great chip for Qualcomm. But they may not. I pay not attention to marketing statements. Let’s see the silicon.

 

[JMacHack]

Well WSL2 was an improvement over WSL and so far MS is supporting and updating WoA.

True but I’m not gonna be convinced until a few years down the road. Maybe I’m too skeptical though.

Full emulation/translation while late considering is here and I think most would argue that the thing holding it back is hardware not OS.

I agree, but that’s a sort of chicken and egg problem.

What is "full-fat ARM support"?

I think Microsoft's current level of commitment to Windows on ARM makes sense.

I mean it seems like a secondary thought, something tacked on to sell a few Qualcomm powered Surface tablets.

I feel like if Microsoft were to believe that ARM support is necessary that they’d iterate much faster in support. With things like x86 emulation becoming faster every annual windows update. Maybe Nuvia will light a fire under their ass, but I can’t imagine it happening with x86 dominance unquestioned.

 

[senttoschool]

I mean it seems like a secondary thought, something tacked on to sell a few Qualcomm powered Surface tablets.

Rewriting and maintaining Windows to work on ARM and emulating x86 on ARM is incredibly expensive. As a software developer, it wouldn't surprise me if this project cost Microsoft hundreds of millions of dollars.

Why would they do this just to sell a few Qualcomm-powered Surface tablets?

 

[Xiao_Xi]

Will Chromebooks/ Linux-based laptops benefit first from Nuvia SoCs?

What steps should Microsoft take to transition to ARM?

 

[Unregistered 4U]

That, to me, seems like a mistake. In the mobile space, Apple as the primary point of comparison makes sense. But while both Apple and Qualcomm will be making ARM-SOCs for the computer, the majority of customers that Qualcomm will be looking to acquire are going to be from the Windows-x86 side. For one thing, it’s a much bigger pool of people. I’d think they’d want to focus on giving those customers a reason to switch from x86 to their ARM cores.

Well, Apple is saying, “We have market leading chips.” Intel, by their comparison to Apple, is saying, “Apple has market leading chips and ours compare well against them.” So, with that in mind, it makes sense that they’d compare themselves to Apple rather than Intel.

Intel’s already laid the groundwork informing the world of how well the M1 performs. They could just advertise that they have processors that perform somewhere in the range of the M1 and they’re in good company.

 

[crazy dave]

Well, Apple is saying, “We have market leading chips.” Intel, by their comparison to Apple, is saying, “Apple has market leading chips and ours compare well against them.” So, with that in mind, it makes sense that they’d compare themselves to Apple rather than Intel.

Intel’s already laid the groundwork informing the world of how well the M1 performs. They could just advertise that they have processors that perform somewhere in the range of the M1 and they’re in good company.

I suppose - especially in the ultraportable range that makes sense. It just seems like an odd focus though when I would’ve thought their target audience would be Windows-x86 users. But maybe it still makes sense.

 

[Unregistered 4U]

That is a false dichotomy. I own an ARM based machine. I did not lose the access to the x86 software ecosystem. Apple has demonstrated that it is possible to run x86 software on an ARM-based machine without significant drawbacks. If other vendors follow them in adopting x86 compatibility features, legacy software support won't be an issue.

More importantly, “losing access to x86” is of little importance to someone buying a new system tomorrow and all they plan to do is edit photos, send emails, surf the web and follow social media.

 

[Unregistered 4U]

I suppose - especially in the ultraportable range that makes sense. It just seems like an odd focus though when I would’ve thought their target audience would be Windows-x86 users. But maybe it still makes sense.

There are more and more Windows-x86 ultraportable systems shipping everyday, with Gartner predicting that marketshare for those will continue to rise as desktop share falls. It’ll be a very important segment that Qualcomm “could” obtain a decent share of if they can ship in reasonable quantities with reasonable performance. Big if

 

[Krevnik]

Rewriting and maintaining Windows to work on ARM and emulating x86 on ARM is incredibly expensive. As a software developer, it wouldn't surprise me if this project cost Microsoft hundreds of millions of dollars.

Why would they do this just to sell a few Qualcomm-powered Surface tablets?

The estimate might be pretty close, but partly because of the long time span of the WoA project (which includes the Surface RT and Windows Phone 8 a decade ago). The fact that it was shelved and then unshelved does mean there was some period of time where nothing was happening with it.

But to get to your question, I think Microsoft is doing this to experiment with ARM on the desktop (versus trying to craft an iPad “killer” with the RT). If Microsoft really wanted to pursue this project with more speed, they could. The fact that they aren’t says that they aren’t fully certain where they should commit resources, how much to commit, or that the market is ready. But it is clear they are interested.

 

[Bodhitree]

I would think people at Microsoft are salivating at the thought of a Qualcomm chip with M1 levels of performance. It would make a serious dent in the x86 business laptop market as long as Office runs on it, just by virtue of superior battery life. I guess we will have to see what the Alder Lake laptops offer for battery life under sustained load.

But I am wondering whether the M1 has moved the needle very much on getting people to shift to the Arm platform. According to the latest sales figures, Mac sales in 2021 were up something like 25% year on year, which I would put down more to Mac afficionados filling in long-delayed purchases rather than Windows users switching in droves to the new hardware.

It is hard to provide computer users with a compelling reason to switch. More speed, more battery life, quiet and cool operation are not huge reasons compared to not being able to game, for students and enthusiasts.

 

[huge_apple_fangirl]

I would think people at Microsoft are salivating at the thought of a Qualcomm chip with M1 levels of performance. It would make a serious dent in the x86 business laptop market as long as Office runs on it, just by virtue of superior battery life. I guess we will have to see what the Alder Lake laptops offer for battery life under sustained load.

Microsoft already dominates the x86 business market. Why should they care what ISA the chip has as long as it runs Windows? Why is there this widespread belief that PCs are on the verge of switching to ARM? The idea that PCs will sacrifice years of backwards compatibility for a marginally better product is misplaced. Caring about the product is Apple's job.

But I am wondering whether the M1 has moved the needle very much on getting people to shift to the Arm platform. According to the latest sales figures, Mac sales in 2021 were up something like 25% year on year, which I would put down more to Mac afficionados filling in long-delayed purchases rather than Windows users switching in droves to the new hardware.

Mix of both.

It is hard to provide computer users with a compelling reason to switch. More speed, more battery life, quiet and cool operation are not huge reasons compared to not being able to game, for students and enthusiasts.

Exactly. The places where ARM has had the most success- Mac and Chrome OS- are platforms where backwards compatibility has never been the selling point.

 

[JMacHack]

More importantly, “losing access to x86” is of little importance to someone buying a new system tomorrow and all they plan to do is edit photos, send emails, surf the web and follow social media.

I think it’s time we put this strawman to bed, I’d wager there are very few people who have only these needs anymore. And the people who do have a smartphone or tablet already. This consumer space is rooted in the 2000’s, and is hardly relevant anymore.

I’d wager the widest consumer of laptop/desktop pcs have computing needs above what can be offered by smartphones or tablets.

To mind, I think the “average full PC consumer” should be defined as someone who: browses the internet/social media, plays video games, is a user of voice and or video chat, and is often a student.

Thus, x86 compatibility is important for things like playing games and using terrible outdated school software. And programs like teamspeak, discord, and mumble aren’t arm native yet.

And let’s not forget the internet isn’t lightweight anymore, even being on youtube or video streaming sites can bring a system to its knees. Raw power matters, even at the cost of wattage.

 

[JMacHack]

But I am wondering whether the M1 has moved the needle very much on getting people to shift to the Arm platform.

I’d say it’s both validated the idea and generated hype. Just going by marketing, the M1 series has been the de facto target in the marketing campaigns of AMD, Intel and NVidia.

I know that before the M1 came out, I was expecting a chip with a marginal performance uplift compared to that gen Intel, with better efficiency. I didn’t expect it to thrash things like it did. And I don’t think anyone else did either.

And as soon as it came out, other companies (not x86) made statements that they were working on competitors to the M1. That’s significant.

According to the latest sales figures, Mac sales in 2021 were up something like 25% year on year, which I would put down more to Mac afficionados filling in long-delayed purchases rather than Windows users switching in droves to the new hardware.

I think that’s correct, though anecdotally I do see a lot of “vocal switchers”.

It is hard to provide computer users with a compelling reason to switch. More speed, more battery life, quiet and cool operation are not huge reasons compared to not being able to game, for students and enthusiasts.

Very much agreed. I think the current PC market is driven by those two demographics. However, this also sells the M1 series short, since they’re capable of playing quite a few games, (just not as wide a library as Windows but that’s always been true) as well as being quality machines. That plus the appeal MacOS has always had (stability, UNIX core, doesn’t have an eye gouging UI, etc) makes them compelling for a wide variety of people.

 

[JMacHack]

Microsoft already dominates the x86 business market. Why should they care what ISA the chip has as long as it runs Windows?

You’re right, they don’t care what ISA their software is running on, but that’s also an argument to keep ARM development up as well. Seeing so many chip companies now looking to target M1 performance has to at least justify some investment, no?

Why is there this widespread belief that PCs are on the verge of switching to ARM? The idea that PCs will sacrifice years of backwards compatibility for a marginally better product is misplaced.

Maybe not this year, maybe not in 5, maybe not ever, but you have to be forward thinking. Maybe one day another company will come up with an ARM chip that whups x86. Then another, etc. Apple has made the shift by making Rosetta 2, sacrificing some performance for a translation layer. If another company brings a killer ARM chip to the table, then that justifies software translation over bare metal compatibility.

Likewise, many apps are made in electron or made to be platform-agnostic. x86 compatibility would matter less than platform compatibility.

I know this argument relies on a lot of “ifs”, but when you have resources like Microsoft, it makes sense to have as much ammunition in your belt as possible.

Exactly. The places where ARM has had the most success- Mac and Chrome OS- are platforms where backwards compatibility has never been the selling point.

Today, yes. Tomorrow, who knows? We can’t just assume that x86 will always remain king of the heap because of bare compatibility.

Today, x86 compatibility is king. But that’s no guarantee for the future.

 

[Bodhitree]

Exactly. The places where ARM has had the most success- Mac and Chrome OS- are platforms where backwards compatibility has never been the selling point.

You fail to mention the elephant in the room — smartphones and tablets. The success of ARM on smartphone has led to the M1 and its success on Mac is the first widespread deployment in mainstream computing. Apple has been the company that made the call that an ARM SoC with unified memory was going to be a better platform for them than the standard PC architecture.

Now the question is, is that also going to hold true in other computing companies? Apple controls the full technology stack that they use, including the OS. With Microsoft, they are trying to keep a version on Intel while at the same time moving their technology to ARM, they have a finger in many other product lines. Are Microsoft going to leverage similar advantages from a Nuvia-designed SoC as Apple did from M1? For example the hardware video encoders.

 

[Unregistered 4U]

I think it’s time we put this strawman to bed, I’d wager there are very few people who have only these needs anymore. And the people who do have a smartphone or tablet already. This consumer space is rooted in the 2000’s, and is hardly relevant anymore.

Undoubtedly, the tablet market (I guess really the iPad market ) is huge and getting larger, but that’s an even better point. Apple sells more iPads in a year than other vendors sell of their entire laptop lines, so that’s another VERY large number of people for whom “access to x86 legacy apps” are of no consequence.

Interestingly, I don’t think we’ll be putting the strawman to bed because your list “browses the internet/social media, plays video games, is a user of voice and or video chat” matches mine pretty closely where performance is concerned. My list has editing photos, yours has gaming, voice and video chat, and there’s nothing in the list that couldn’t be provided in a non-x86 package (it’s being done today for the iPad)… especially if the developers of those packages were aware that Microsoft and it’s OEM’s would soon be releasing a product they MIGHT want to get their code working on or risk losing users.

Thus, x86 compatibility is important for things like playing games and using terrible outdated school software. And programs like teamspeak, discord, and mumble aren’t arm native yet.

If teamspeak, discord and mumble weren’t currently in Apple’s App Store, there might be cause for concern as that would indicate those platforms weren’t actively being supported. As they appear to be supported, you can bet that they would all have versions ready for release (at least as betas) as soon as the first ARM Windows systems hit the virtual shelves. “x86 compatibility” is mainly required for industries/institutions/organizations where there are mission critical applications that are no longer being developed/supported and there’s no possibility of finding an alternate solution. Those are few and getting fewer. For everyone using fairly recent software still being supported by dev teams, those dev teams are going to support where their users go.

And let’s not forget the internet isn’t lightweight anymore, even being on youtube or video streaming sites can bring a system to its knees. Raw power matters, even at the cost of wattage.

Yes, but “performance” isn’t synonymous with “access to x86 legacy apps”. There are many examples of systems that are performant enough for today’s internet that aren’t compatible with x86.

 

[mi7chy]

Will Chromebooks/ Linux-based laptops benefit first from Nuvia SoCs?

What steps should Microsoft take to transition to ARM?

Mediatek is a better fit for Chromebooks since one of the key points is value like the $200'ish Lenovo Duet Chromebook that offers a full desktop browser. Once you get into the $400+ range you're better off just getting a Windows x64 device.

There are already a handful of Windows on ARM devices but it's a tough niche. From my usage, Windows on ARM is more polished than ChromeOS and MacOS but won't gain traction against Windows AMD x64 with equivalent battery life but much more versatility and often at lower prices. For this segment to grow it needs more $200 to 300'ish devices like the Lenovo Duet but running Windows on ARM.

Acer Spin 7
HP Elite Folio
Lenovo Flex 5G
Microsoft Surface Pro X
Samsung Galaxy Book S
Samsung Galaxy Book Go

 

[JMacHack]

Undoubtedly, the tablet market (I guess really the iPad market ) is huge and getting larger, but that’s an even better point. Apple sells more iPads in a year than other vendors sell of their entire laptop lines, so that’s another VERY large number of people for whom “access to x86 legacy apps” are of no consequence.

Interestingly, I don’t think we’ll be putting the strawman to bed because your list “browses the internet/social media, plays video games, is a user of voice and or video chat” matches mine pretty closely where performance is concerned. My list has editing photos, yours has gaming, voice and video chat, and there’s nothing in the list that couldn’t be provided in a non-x86 package (it’s being done today for the iPad)… especially if the developers of those packages were aware that Microsoft and it’s OEM’s would soon be releasing a product they MIGHT want to get their code working on or risk losing users.


If teamspeak, discord and mumble weren’t currently in Apple’s App Store, there might be cause for concern as that would indicate those platforms weren’t actively being supported. As they appear to be supported, you can bet that they would all have versions ready for release (at least as betas) as soon as the first ARM Windows systems hit the virtual shelves. “x86 compatibility” is mainly required for

I didn’t realize they were in the App Store, huh.

industries/institutions/organizations where there are mission critical applications that are no longer being developed/supported and there’s no possibility of finding an alternate solution. Those are few and getting fewer. For everyone using fairly recent software still being supported by dev teams, those dev teams are going to support where their users go.

I agree with this, except big monetary institutions seem to be hell bent on keeping old software. (Which is why COBOL devs exist and make beaucoup bux) just an example.

Yes, but “performance” isn’t synonymous with “access to x86 legacy apps”. There are many examples of systems that are performant enough for today’s internet that aren’t compatible with x86.

Naturally, I agree. But I’m just saying a certain level of performance is expected even by the basic consumer.

 

[Unregistered 4U]

I agree with this, except big monetary institutions seem to be hell bent on keeping old software. (Which is why COBOL devs exist and make beaucoup bux) just an example.

Yes, and no one can deny this. The big monetary institutions will be the last to do anything significant because they know that THEIR money/tools outweighs a user’s OS preference EVERY time. I’d even bet some large tooling manufacturers, (as they retire old machines and acquire new ones) will be off of legacy x86 well before the financial institutions!

Naturally, I agree. But I’m just saying a certain level of performance is expected even by the basic consumer.

Oh, absolutely. I just have little doubt that non-x86 systems would be able to bring that level of performance

 

[Gerdi]

I mean it seems like a secondary thought, something tacked on to sell a few Qualcomm powered Surface tablets.

I feel like if Microsoft were to believe that ARM support is necessary that they’d iterate much faster in support. With things like x86 emulation becoming faster every annual windows update. Maybe Nuvia will light a fire under their ass, but I can’t imagine it happening with x86 dominance unquestioned.

Windows ARM is a full featured Windows including x86 and x64 emulation, WSL2 etc. Not even sure what you mean with "iterate much faster". In addition to Windows itself, Microsoft was the driving force of bringing major frameworks natively to Windows ARM like OpenJDK and .Net including WPF and Winforms. WPF and Winforms support was even backported to .Net Core 5.

 

[deconstruct60]

It seems that when Nuvia-based SOCs finally release for PC the primary comparison Qualcomm will be drawing against will be Apple’s M-series SOCs rather than Intel/AMD chips.

That, to me, seems like a mistake. In the mobile space, Apple as the primary point of comparison makes sense.

Qualcomm isn't talking about anything but the mobile space. Apple is the one running around talking about what they have beats all desktops. I don't think Qualcomm is trying to assume Apple's smack talking in taking the position of using Apple's M1 as a metric. [ Qualcomm is not any anyway shape or form really talking about taking on the M1 Max class SoC and probably a slippery slope they are talking about a M1 Pro class SoC ]

About four factors are highly relevant . First, Qualcomm has already made statements about walking past x86 already before even got to the Nuvia solution. About "7:20 pm" in the timeline in they Tech Day briefing in early December 2021 on the 8cx Gen 3 .

Qualcomm Snapdragon Tech Summit: Day 2 Keynote on ACPC and Gaming

www.anandtech.com

Yes, this is before the Intel Gen 12 ( Alder Lake ) U series or Ryzen 6000U series benchmarks were out, but it is indicative that just with Arm's X-series progression Qualcomm already has a foundation to put x64_86 solutions on a back foot. They don't need Nuvia to beat x86_64.

The 8cx Gen has 4 X1s and 4 x A78s to move past x86_64.

Qualcomm’s 8cx Gen 3 for Notebooks, Nuvia Core in 2022/2023

www.anandtech.com

They haven't even gotten to X2 and A710. Or to X3 and whatever next iteration is for mid-side A
Even if Nuvia core implodes , Qualcomm essentially has a Plan B. that will keep them in the ultramobile game.


Qualcomm has another problem. MediaTek has already got an X2 phone version.

MediaTek Announces Dimensity 9000: Supercharged Flagship SoC on 4nm

www.anandtech.com

Adjust that 1 x X2 + 3 x A710 + 4 x A510 to a 4 x X2 + 4 x A710 and bulk up the GPU and they are extremely competitive with the ultramobile x86_64 solutions also. It is a matter of pricing the SoC.

So if Qualcomm gets into a position of our solutions are closer to M1 than Mediatek, Samsung+AMD/ etc. that will actually be a win at being the Windows on Arm leading solution. If Nuvia is a bust then they had to 'win' on iGPU. With AMD partnering up with Samsung and on Windows ... that is much thinner ice.


Second, Apple's M-series is much more not solely focus on CPU cores. There is a pretty hefty weighting in the M1 ( and even higher in M1 Pro) on the iGPU. Nuvia isn't really going to solve that by itself. If they had a high bandwidth , low latency interconnect and memory bandwidth approach in place that probably could help Qualcomm walk their iGPU into a 2020+ era laptop competitive field. x86_64 ultramobile not so sure they are the best foundational benchmark to fully ( especially when it comes to memory bandwidth to iGPU. )

Pretty good chance that Qualcomm doesn't try to get into a core count war with the high end mobile (and desktop) x86_64. Like the M1 Pro may not see the "even" big.little that they have done with the Arm reference modified designs. So 4-6 x X2 + 2 x A710 and allocate saved space to iGPU or to fixed function.

[ If Qualcomm wanted to get into a head-to-head with Apple, then I think they would have to loose the Celluar modem part. Qualcomm has had a problem in attempts up to the present in that they have wanted to charge a huge price premium for being "always on" connected. Users wanted more battery life and no drop in performance more than "always on". Or Qualcomm is expecting Apple to couple its modem to the plain Mx and level the playing field for them. ]



Three , For laptops ( and certainly for desktops ) they don't necessarily need a A5x class core. It could be nice to have. But Windows would probably do just fine with a A7x class core as the 'E' cores. The x86_64 solutions setting that up ( Intel is mid-size and AMD has no 'E' cores for a substantive amount of time into future and pretty likely to just follow Intel's approach when they do. )

There is pretty decent chance Nuvia can salvage what they were going to use for their server core if it was tweaked for very high core counts on a die. Converting that into a "mid-size" E core probably could be done with some adjustments. It just has to be more area optimized and power optimized than the "bigger" core. If they were shooting for putting 64-128 cores on a mid-size die then they probably already have that design to a large extent ( may have to strip out some aggressive SIMD/Vector stuff but the other function units could have been pretty close).


Four, Pricing. Seems doubtful Qualcomm will want to get into the large die or multiple chip module end of the market. Pragmatically that means they wouldn't be chasing desktop products. Like Apple, they'll probably be underprovision relative to dGPU PCI-e lane use cases. High end laptop AMD/Intel SoC will likely continue to provision to promote synergy sales of dGPUs. Can look at the rest of that interview

"... The way we look at it is that we're not going after discrete designs: we're going after an SoC and the architecture that makes the best sense for the PC. ... "

That is an addition motivation to switching away from doing heavy comparisons to x86_64 if those vendors are significantly focused on enabling dGPUs. They aren't going to be talking about the same subset of the overall PC market.

If Apple is kicking dGPUs 'to the curb' , then so it Qualcomm so that gets to be more of an "apples to apples" comparison point over the long term. They can avoid the tech porn press that will be circling around how can hook the Nvidia 5090M up to their SoC in the future.

I think it is a big mistake to read " for the PC" and map that into the entirety of the mainstream Windows market. It isn't. They are shooting for a defined subset. Just like Apple is shooting for a defined subset. However, even those are not the same subset. ( Apple has more desktop ambitions weaved in... they have essentially explicitly said so. )

But while both Apple and Qualcomm will be making ARM-SOCs for the computer, the majority of customers that Qualcomm will be looking to acquire are going to be from the Windows-x86 side. For one thing, it’s a much bigger pool of people. I’d think they’d want to focus on giving those customers a reason to switch from x86 to their ARM cores.

The pool of ultraprotable and under $1,100 Windows x86+64 market is still bigger than the overall Mac market. Qualcomm doesn't have to pentrate out of the laptop market to be bigger in volume than what Apple is selling.

Maybe they think their new SOCs will be so good that they won’t have to draw comparisons with AMD/Intel. But that’s not the answer he gave (despite Ian leaving the door open for that).

They already did that this past December. Even the Arm reference design that don't sacrifice area for performance are winning... custom or reference design they have something. That part isn't "interesting" anymore.

When Qualcomm gets some Windows on Arm competitors with shipping SoCs, I'm pretty sure the comparisons will shift over to them. For the immediate future though Apple is the more natural "aspirational" point.

 

[Xiao_Xi]

How good is the Snapdragon 8cx Gen 3?

 

[Bodhitree]

I feel Nuvia’s chip is likely to come in as a low-cost alternative for cheap laptops, chromebooks and Android tablets. It would have to be something pretty special to move the top-end of the PC market given that most software would be running under emulation, and the PC enthusiast market is rather sensitive to framerates.

 

[deconstruct60]

He’s referring to Nuvia which was bought by Qualcomm. While the individuals involved are all highly experienced ex-Apple and Google chip designers, @cmaier is dubious as Nuvia was bought before bringing a product to market so has not yet backed up its claims with shipping silicon.

But if read the interview Nuvia isn't going to do the whole SoC.

"... By that I mean that it's impossible for us to put out a chipset solution as sophisticated as this without having the entire system being taken into consideration. Think of it this way: the CPU by itself is part of the ‘one technology roadmap’, but so is graphics, and other things. Then we're really thinking about bringing a complete system solution to the PC and changing it in such a way that you don't go after the traditional designs. ... "

One of the tricky parts here is how quickly they can weave the Nuvia work in with the rest of the SoC work. IMHO, that is at least as big of a negative potential hiccup here as the CPU core design "failing" or "stumbling". Whose memory subs systems are they using? Interconnect?

On one hand it could be that the Nuvia folks now have to do substantially less. (.e.g, just CPU cores that integrates to an existing interconnect. Fewer 'distractions' better chance to get the kinks out ). On other hand it trying to pull the other Qualcomm stuff onto a foundation that hasn't deployed in the field before ... that could get into a fubar state quickly.

If Qualcomm has folks doing the graphic cores implementation optimizations and AI/ML core implementation optimizations, then it isn't like the Nuvia folks would be lost in woods with no one internal to help them with questions or give 'heads up' on potential issues. The fabrication challenges can vary somewhat from CPU , GPU , NPU/AI , DSP implementation but it isn't completely decoupled either. memory caches even less so between function units.

That said. Sampling in 2022 and shipping in 2023 I doubt this first iteration will be a "maximally" optimized solution. Getting it working is probably higher priority than squeezing all the performance possible out of it or having to exactly made the 'slide ware' that Nuvia had when hunting for investors.

I wouldn't be surprised if the Nuvia cores were less area and power optimized than what Qualcomm needs for the smartphone space, but decent for the less requirements constrained PC laptop space. Then on the 2nd generation they move to tighter constraints. That way they can iterate down as the integration between the two companies matures.

Similarly skip anything like AMX ( Apple or Intel matrix module ) or trying to leapfrog on some AI/ML modification attached to the CPU core complex. That could offset some area expansion needed to be a bit more aggressive.

I think at best there will be some cherry picked microbenchmarks where they might get close to M1 but it will be something with more rough edges. Which in the context ( done in the middle of a company and design team integration) would be pretty good. Not sure why anyone would be expecting some amazing grand slam homerun here.. Just incremental improvement over the 8cx Gen 3 would be a significant feat development logistics wise .

 

[crazy dave]

Qualcomm isn't talking about anything but the mobile space. Apple is the one running around talking about what they have beats all desktops. I don't think Qualcomm is trying to assume Apple's smack talking in taking the position of using Apple's M1 as a metric. [ Qualcomm is not any anyway shape or form really talking about taking on the M1 Max class SoC and probably a slippery slope they are talking about a M1 Pro class SoC ]

About four factors are highly relevant . First, Qualcomm has already made statements about walking past x86 already before even got to the Nuvia solution. About "7:20 pm" in the timeline in they Tech Day briefing in early December 2021 on the 8cx Gen 3 .

Qualcomm Snapdragon Tech Summit: Day 2 Keynote on ACPC and Gaming

www.anandtech.com

Yes, this is before the Intel Gen 12 ( Alder Lake ) U series or Ryzen 6000U series benchmarks were out, but it is indicative that just with Arm's X-series progression Qualcomm already has a foundation to put x64_86 solutions on a back foot. They don't need Nuvia to beat x86_64.

The 8cx Gen has 4 X1s and 4 x A78s to move past x86_64.

Qualcomm’s 8cx Gen 3 for Notebooks, Nuvia Core in 2022/2023

www.anandtech.com

They haven't even gotten to X2 and A710. Or to X3 and whatever next iteration is for mid-side A
Even if Nuvia core implodes , Qualcomm essentially has a Plan B. that will keep them in the ultramobile game.


Qualcomm has another problem. MediaTek has already got an X2 phone version.

MediaTek Announces Dimensity 9000: Supercharged Flagship SoC on 4nm

www.anandtech.com

Adjust that 1 x X2 + 3 x A710 + 4 x A510 to a 4 x X2 + 4 x A710 and bulk up the GPU and they are extremely competitive with the ultramobile x86_64 solutions also. It is a matter of pricing the SoC.

So if Qualcomm gets into a position of our solutions are closer to M1 than Mediatek, Samsung+AMD/ etc. that will actually be a win at being the Windows on Arm leading solution. If Nuvia is a bust then they had to 'win' on iGPU. With AMD partnering up with Samsung and on Windows ... that is much thinner ice.


Second, Apple's M-series is much more not solely focus on CPU cores. There is a pretty hefty weighting in the M1 ( and even higher in M1 Pro) on the iGPU. Nuvia isn't really going to solve that by itself. If they had a high bandwidth , low latency interconnect and memory bandwidth approach in place that probably could help Qualcomm walk their iGPU into a 2020+ era laptop competitive field. x86_64 ultramobile not so sure they are the best foundational benchmark to fully ( especially when it comes to memory bandwidth to iGPU. )

Pretty good chance that Qualcomm doesn't try to get into a core count war with the high end mobile (and desktop) x86_64. Like the M1 Pro may not see the "even" big.little that they have done with the Arm reference modified designs. So 4-6 x X2 + 2 x A710 and allocate saved space to iGPU or to fixed function.

[ If Qualcomm wanted to get into a head-to-head with Apple, then I think they would have to loose the Celluar modem part. Qualcomm has had a problem in attempts up to the present in that they have wanted to charge a huge price premium for being "always on" connected. Users wanted more battery life and no drop in performance more than "always on". Or Qualcomm is expecting Apple to couple its modem to the plain Mx and level the playing field for them. ]



Three , For laptops ( and certainly for desktops ) they don't necessarily need a A5x class core. It could be nice to have. But Windows would probably do just fine with a A7x class core as the 'E' cores. The x86_64 solutions setting that up ( Intel is mid-size and AMD has no 'E' cores for a substantive amount of time into future and pretty likely to just follow Intel's approach when they do. )

There is pretty decent chance Nuvia can salvage what they were going to use for their server core if it was tweaked for very high core counts on a die. Converting that into a "mid-size" E core probably could be done with some adjustments. It just has to be more area optimized and power optimized than the "bigger" core. If they were shooting for putting 64-128 cores on a mid-size die then they probably already have that design to a large extent ( may have to strip out some aggressive SIMD/Vector stuff but the other function units could have been pretty close).


Four, Pricing. Seems doubtful Qualcomm will want to get into the large die or multiple chip module end of the market. Pragmatically that means they wouldn't be chasing desktop products. Like Apple, they'll probably be underprovision relative to dGPU PCI-e lane use cases. High end laptop AMD/Intel SoC will likely continue to provision to promote synergy sales of dGPUs. Can look at the rest of that interview

"... The way we look at it is that we're not going after discrete designs: we're going after an SoC and the architecture that makes the best sense for the PC. ... "

That is an addition motivation to switching away from doing heavy comparisons to x86_64 if those vendors are significantly focused on enabling dGPUs. They aren't going to be talking about the same subset of the overall PC market.

If Apple is kicking dGPUs 'to the curb' , then so it Qualcomm so that gets to be more of an "apples to apples" comparison point over the long term. They can avoid the tech porn press that will be circling around how can hook the Nvidia 5090M up to their SoC in the future.

I think it is a big mistake to read " for the PC" and map that into the entirety of the mainstream Windows market. It isn't. They are shooting for a defined subset. Just like Apple is shooting for a defined subset. However, even those are not the same subset. ( Apple has more desktop ambitions weaved in... they have essentially explicitly said so. )





The pool of ultraprotable and under $1,100 Windows x86+64 market is still bigger than the overall Mac market. Qualcomm doesn't have to pentrate out of the laptop market to be bigger in volume than what Apple is selling.




They already did that this past December. Even the Arm reference design that don't sacrifice area for performance are winning... custom or reference design they have something. That part isn't "interesting" anymore.

When Qualcomm gets some Windows on Arm competitors with shipping SoCs, I'm pretty sure the comparisons will shift over to them. For the immediate future though Apple is the more natural "aspirational" point.

Just to clarify by the mobile space in that first post I meant iPhone/Android mobile. There Apple not only has mind share but a significant market fraction. So targeting that to actually get people to switch makes sense. Less so for the ultraportables. In the ultraportables, Apple has mindshare but not quite the same level of market share (although ultraportables are their biggest market share I don’t know how big it is). I agree that it can still be useful to compare as Apple has set themselves up as having the best ultraportable around (mindshare) but my concern is that they are risking not convincing the customers of x86 Windows ultraportables to become ARM Windows ultraportable customers. This strategy might be viable if in the minds of those customers they think “I’d get an air if only it ran Windows and here is something as good or better”. Very possible. But they might run into the problem of “yeah I wouldn’t get the air either. How do I know my software will run?”.

While the current ARM designs may have “something” to recommend them, the devices are overall not great and I don’t see having a great reputation. You can tell by how slowly if at all software has been ported and the limited selection of software. Part of that is because full x64 emulation took so long. But a lot of that comes down to the hardware - Qualcomm has half-assed the chips in my opinion. They just aren’t good value. LTT did an entire video on “how did MS screw this up to the point that I forgot these existed?” when discussing WoA. And the Qualcomm CEO had to point to the launch of the M1 to say “see? this could work!” Not one of his own devices.

So if these forthcoming Nuvia chips are really good, then I actually think comparisons with x86 ultraportable is still warranted, even vital - particularly when running x86 software vs native. That’s going to be an important point for their pool of customers. That’s why I think that’s if I were Qualcomm I’d yes talk up how much better we are than the air, “we’re now the best ultraportable” but primarily I’d be telling Windows users that they’ll get to run all their software and it will feel just as good if not better because our new chips are so often. That, to me, is how you ship more Windows ultraportables to Windows users than Apple ships ultraportables to Mac users.