Discussion Qualcomm Snapdragon Thread

[eek2121]

I disagree. I think the key mainstream metric is getting through the day. “Oh I forgot to bring my charger, will it last until I get back.” The AMD is doing that for most people. Not that the greater battery won’t turn some to the Qualcomm, but the impact past one day is not nearly as great.

AMD and Intel clearly have a lot of work to do, they can’t sit on this, but neither are in danger of a marketshare collapse.

AMD is in no danger lol.

They target what matters. If x86 were doomed, Apple would have killed it. They didn’t.

AMD also has an ARM license and could launch their own ARM chip if they chose to. They already use it in Ryzen and EPYC.

Qualcomm’s biggest hurdles are:
1) They are qualcomm. They love margins more than AMD.
2) Compatibility with hardware and software is *still* a nonstarter for most.
3) They are trying to be a 1 stop shop instead of working with others to build a robust ARM platform. Example: This means they are building their own GPU and everything instead of letting NVIDIA do it. They could have simply partnered with NVIDIA, but nope!
3) They don’t have a server platform and client isn’t expandable/upgradeable.

Hopefully they will improve, but I am not holding my breath. PCIE and DDR5 slots would be a great start.

Again, I am not against ARM, I just don’t have faith in Qualcomm to deliver. For them to deliver they would need to completely change as a company, possibly even sell chips at a loss to gain marketshare.

Sure, battery life can be great, but that doesn’t matter if you can’t run the software and hardware you own.

My mouse and keyboard software don’t work, printer. Sound card. GPU. RAM. USB devices? hit or miss. At a previous job we had scan guns that were programmed via USB. No drivers on ARM. The CRM they use has drivers for interfacing with the phone system. Does not work on ARM. The vendor in both cases has no plans to fix. Vendor 1 is a 20 person company and doesn’t have the resources since in their eyes they make hardware, not software. Vendor #2 just doesn’t care because all their clients use x86.



This isn’t meant to be a rant, I just wanted to point all this out.

 

[Doug S]

We are witnessing a renaissance of custom ARM core designs.

I guess ARM the company isn't very happy about this, because ALAs have lower royalty rates than TLAs.

When Nvidia was looking to buy ARM at $60 billion or whatever that crazy price was I said there was no possible way unless they massively increased their royalty rates, and that all that would do is drive companies toward custom cores to reduce their royalty payments. ARM's market cap is now more than double that figure, and all they can talk about is how they are increasing their royalty rates. So this outcome was inevitable.

The lower volume companies who can't justify the cost of designing their own might become customers of RISC-V down the road, depending on how tightly ARM squeezes. There just isn't any way ARM can justify anything remotely close to a $100 billion valuation now or ever. The big fish can design their own cores and ignore ARM's, and the small fish making embedded stuff or whatever don't have any need for ARM beyond the massive infrastructure that has been built up around it in the past 25 years. That infrastructure is starting to be replicated in the RISC-V world.

 

[igor_kavinski]

That infrastructure is starting to be replicated in the RISC-V world.

But it will end up becoming the Linux of ISAs. Not entirely a great thing. Fragmentation fun!

 

[lakedude]

That's jut corporate speak for:

1) Make a product
2) Waves hands >poof< millions of sales
3) Profit!

 

[SpudLobby]

When Nvidia was looking to buy ARM at $60 billion or whatever that crazy price was I said there was no possible way unless they massively increased their royalty rates, and that all that would do is drive companies toward custom cores to reduce their royalty payments. ARM's market cap is now more than double that figure, and all they can talk about is how they are increasing their royalty rates. So this outcome was inevitable.

The lower volume companies who can't justify the cost of designing their own might become customers of RISC-V down the road, depending on how tightly ARM squeezes. There just isn't any way ARM can justify anything remotely close to a $100 billion valuation now or ever. The big fish can design their own cores and ignore ARM's, and the small fish making embedded stuff or whatever don't have any need for ARM beyond the massive infrastructure that has been built up around it in the past 25 years. That infrastructure is starting to be replicated in the RISC-V world.

Yeah a fundamental problem with Arm is this.

You don’t in fact need world-class CPU design validation and support teams indefinitely after building a great ISA and maintaining it. The ISA should be somewhat of a consortium or regardless a firm split off from any competing business in developing CPUs with said ISA for license. It’s still more open than X86, but it’s clearly a nonsensical legacy. It’s possible at some point with Arm’s software base growing we see Samsung/Nvidia/Qualcomm/Apple (yes Inb4 Apple has special privileges but who knows) et. Al pay ransom to split it. It’s just silly and it’s already a technically near-perfect ISA for modern broad high performance use*

*SVE garbage aside

 

[Doug S]

But it will end up becoming the Linux of ISAs. Not entirely a great thing. Fragmentation fun!

For embedded that doesn't matter. You buy a finished core design (or finished SoC) for your project that includes what you need, and you compile your Linux kernel, write your custom code, etc. for it. If your core doesn't include SIMD instructions because you didn't spec them as a requirement who cares, that just means you give a flag to your compiler to not generate them.

The only market of any size where your customers are downloading binaries that weren't compiled with the vendor's toolchain is Android, but companies making wireless routers or security cameras or drones and on and on are not running Android and don't care about what toolchain might have been used for apps on the Google Play store.

There is fragmentation on ARM too if you look at embedded. Just look at the differences between the various M core families.

 

[Nothingness]

For embedded that doesn't matter. You buy a finished core design (or finished SoC) for your project that includes what you need, and you compile your Linux kernel, write your custom code, etc. for it.

When you buy an embedded core design, it will get uses beyond the initial design target/market; from a software point of view, your design might have to support more recent versions of SW stacks, kernel included. I did hit this when I was porting SW to PS2 Linux; Sony/Toshiba provided obsolete tools for the MIPS R5900 and I had to port binutils/gcc changes to more recent versions, which ended costing me much more time than the SW porting itself. I endured something similar when upgrading my Linux-based router firmware.

As soon as your CPU isn't standard and your platform allows software upgrades having to rely on vendor specific tools and software is a pain.

Interestingly both the platforms that were causing me pains were using some variants of MIPS ISA. RISC-V is reproducing the same hell.

That being said, fragmentation exists on Arm platforms too and not only for M-class CPUs, but there the number of vendor specific extensions is very small.

If your core doesn't include SIMD instructions because you didn't spec them as a requirement who cares, that just means you give a flag to your compiler to not generate them.

And you pay the cost of SW validation multiple times. And you don't use assembly or intrinsics anywhere.

 

[FlameTail]

AMD is in no danger lol.

They target what matters. If x86 were doomed, Apple would have killed it. They didn’t.

AMD also has an ARM license and could launch their own ARM chip if they chose to. They already use it in Ryzen and EPYC.

Qualcomm’s biggest hurdles are:
1) They are qualcomm. They love margins more than AMD.
2) Compatibility with hardware and software is *still* a nonstarter for most.
3) They are trying to be a 1 stop shop instead of working with others to build a robust ARM platform. Example: This means they are building their own GPU and everything instead of letting NVIDIA do it. They could have simply partnered with NVIDIA, but nope!
3) They don’t have a server platform and client isn’t expandable/upgradeable.

Hopefully they will improve, but I am not holding my breath. PCIE and DDR5 slots would be a great start.

Again, I am not against ARM, I just don’t have faith in Qualcomm to deliver. For them to deliver they would need to completely change as a company, possibly even sell chips at a loss to gain marketshare.

Sure, battery life can be great, but that doesn’t matter if you can’t run the software and hardware you own.

My mouse and keyboard software don’t work, printer. Sound card. GPU. RAM. USB devices? hit or miss. At a previous job we had scan guns that were programmed via USB. No drivers on ARM. The CRM they use has drivers for interfacing with the phone system. Does not work on ARM. The vendor in both cases has no plans to fix. Vendor 1 is a 20 person company and doesn’t have the resources since in their eyes they make hardware, not software. Vendor #2 just doesn’t care because all their clients use x86.

You are missing the forest for the trees.

The shrewd people can already see the writing on the wall: ARM will make x86 marketshare bleed.

It's like when Zen launched 8 years ago in 2017. Zen 1 was obviously an amazing architecture, but it wasn't without shortcomings. Some shortsighted pundits used that fact to dismiss AMD/Zen as not being a threat. They couldn't have been more wrong. In the 8 years since, Intel has bled like a butchered pig. Their margins in datacenter have collapsed. Meanwhile, AMD's marketshare in client/server is higher than ever.

Now the same thing is happening again, but this time it's x86 vs ARM.

Sure, Windows-on-ARM has a gazillion compatibility issues and first generation Snapdragon X was a mixed bag. But here's the thing: it's only going to get better from now.

Even Zen 1 was a mixed bag when it launched. It didn't cause AMD to gain +10% marketshare overnight. It took AMD several iterations to iron out the kinks and surpass Intel in all performance/efficiency metrics.

As the recent Semianalysis article notes, there are a plethora of players with deep pockets and/or great engineering capabilities who have a vested interested in making ARM for PCs work.

Yes, Arm for PC still has many kinks to iron out, so Qualcomm’s Snapdragon X hasn’t taken much market share. What’s important is that the dam has broken and a flood will start soon. Arm for PC will happen because there is now a quorum of important players in the ecosystem (Microsoft, Arm, Qualcomm, Nvidia, Mediatek) who want to and are set on making Arm for PC happen.

Intel on the Brink of Death | Culture Rot, Product Focus Flawed, Foundry Must Survive

Intel’s board is incompetent and its horrible decisions over the decades are going to push it towards death. The decision to fire Pat Gelsinger, put in charge a CFO + career sales and marketing lea…

semianalysis.com

Microsoft upgrades Prism with support for AVX, AVX2, BMI, FMA, and F16C Emulation

Qualcomm's Snapdragon X Elite Gen 2 will include third-gen Oryon cores


The Ultimate CPU: Arm Cortex-X925’s Breakthrough with a 15 Percent IPC Improvement


Arm has published its PC Base System Architecture (PC-BSA) specification, the blueprint for standardizing Arm-based client PCs.

Exclusive: MediaTek designs Arm-based chip for Microsoft's AI laptops

Exclusive: Nvidia to make Arm-based PC chips in major new challenge to Intel

//

PS: To be clear, ARM will not make x86 obsolete any time soon. But it will make x86 marketshare bleed, and most of the bleeding will be by Intel. AMD has shown that they are nimble and can adapt to new trends, and their engineering capabilities are also broadly better. The Semianalysis author agrees;

No, x86 will not disappear overnight. It is still a large market and potentially a cash cow business. But cash cow status only happens if large swaths of employees are fired, choking innovation long term. Even then, AMD and the various Arm players likely grab market share faster than the Intel board is thinking. The board’s ”focus on product” strategy sounds like a dead end

 

[eek2121]

You are missing the forest for the trees.

The shrewd people can already see the writing on the wall: ARM will make x86 marketshare bleed.

It's like when Zen launched 8 years ago in 2017. Zen 1 was obviously an amazing architecture, but it wasn't without shortcomings. Some shortsighted pundits used that fact to dismiss AMD/Zen as not being a threat. They couldn't have been more wrong. In the 8 years since, Intel has bled like a butchered pig. Their margins in datacenter have collapsed. Meanwhile, AMD's marketshare in client/server is higher than ever.

Now the same thing is happening again, but this time it's x86 vs ARM.

Sure, Windows-on-ARM has a gazillion compatibility issues and first generation Snapdragon X was a mixed bag. But here's the thing: it's only going to get better from now.

Even Zen 1 was a mixed bag when it launched. It didn't cause AMD to gain +10% marketshare overnight. It took AMD several iterations to iron out the kinks and surpass Intel in all performance/efficiency metrics.

As the recent Semianalysis article notes, there are a plethora of players with deep pockets and/or great engineering capabilities who have a vested interested in making ARM for PCs work.

Intel on the Brink of Death | Culture Rot, Product Focus Flawed, Foundry Must Survive

Intel’s board is incompetent and its horrible decisions over the decades are going to push it towards death. The decision to fire Pat Gelsinger, put in charge a CFO + career sales and marketing lea…

semianalysis.com

Microsoft upgrades Prism with support for AVX, AVX2, BMI, FMA, and F16C Emulation

Qualcomm's Snapdragon X Elite Gen 2 will include third-gen Oryon cores
View attachment 112938

The Ultimate CPU: Arm Cortex-X925’s Breakthrough with a 15 Percent IPC Improvement
View attachment 112937

Arm has published its PC Base System Architecture (PC-BSA) specification, the blueprint for standardizing Arm-based PCs.

//

PS: I do think most of the x86 marketshare bleeding will be by Intel. AMD has shown that they are nimble and can adapt to new trends, and their engineering capabilities are also broadly better.

I think you are. Performance and battery life don’t matter when the compatibility isn’t there. Don’t believe me?

From the surface laptop on Amazon:

 

[soresu]

Arm has published its PC Base System Architecture (PC-BSA) specification, the blueprint for standardizing Arm-based PCs.

*ARM based client PCs.

ARM already began a standard for server systems (SBSA) a decade ago in Jan 2014.

 

[LightningZ71]

While I realize that it's not an issue for the vast majority of the market, my biggest pet peave for ARM laptops is that I still haven't seen any with upgradeable RAM. I know Apple doesn't have this either, but it's still the majority of non-sub $499 laptops on the market that have it. Does anyone know if any of the ARM laptops have upgradeable SSDs either?

 

[soresu]

It's like when Zen launched 8 years ago in 2017. Zen 1 was obviously an amazing architecture, but it wasn't without shortcomings. Some shortsighted pundits used that fact to dismiss AMD/Zen as not being a threat. They couldn't have been more wrong. In the 8 years since, Intel has bled like a butchered pig. Their margins in datacenter have collapsed. Meanwhile, AMD's marketshare in client/server is higher than ever.

To be fair to Intel a lot of their woes stemmed from their CPU/SoC engineering efforts being so tightly coupled to their semiconductor manufacturing division, and the problems incurred when they got overambitious for 10nm and hit that giant roadblock only 2 nodes after crowing their finFET/trigate dominance.

Being so tightly coupled to expected process node targets meant their adaptability to delays was very poor compared to AMD, whose main constraint in process was legal/contractual to GloFo earlier in the Ryzen roadmap.

 

[MS_AT]

PS: To be clear, ARM will not make x86 obsolete any time soon. But it will make x86 marketshare bleed, and most of the bleeding will be by Intel. AMD has shown that they are nimble and can adapt to new trends, and their engineering capabilities are also broadly better. The Semianalysis author agrees;

Sure, they will slowly bleed them out. But I am not sure it will be Qualcomm, I can see NVidia in this role. Maybe M$ doing their own chips, or using custom AMD ones. But let's be serious after 8 years of trying Qualcomm was able to produce one decent SoC, that is already obsoleted by its own follow-up mobile SoC, that is overbuilt for the purpose it's serving.

I mean Qualcomm has a nice opportunity here, make the flagship 8 Elite SoC able to dual-boot Windows/Linux, make some quality rules for the USB-C connection types and provide shells for the smartphone that would be able to turn the device into a desktop or bigger laptop if I needed. Make it possible to hook into eGPU for gaming. That would be something new on the market. Something fresh, that could justify the price premium, if you could keep your daily compute in your pocket and seamlessly transition to desktop.

I mean winning chrome benchmarks is fine, but maybe I am weird but given the chance I prefer to browse on bigger screen with mouse and desktop support. And before you say Qualcomm is only supplying chips, wasn't that true also for Intel, but they could launch all those initiatives like ultrabook, Intel Evo, etc. That would show in my eyes that they are serious about the market. But I am not a business guy, so I may have no idea what I am talking about.

 

[FlameTail]

While I realize that it's not an issue for the vast majority of the market, my biggest pet peave for ARM laptops is that I still haven't seen any with upgradeable RAM. I know Apple doesn't have this either, but it's still the majority of non-sub $499 laptops on the market that have it.

Upgradeable RAM is not possible at the moment because ARM SoCs use LPDDR, which was non-socketable until...

LPCAMM released. But LPCAMM is relatively new, and it's very very expensive. There is only a handful x86 laptops with it. I believe we will see ARM laptpps with socketed RAM in the future, when LPCAMM becomes cheaper and ubiquitous.

Does anyone know if any of the ARM laptops have upgradeable SSDs either?

All Snapdragon X laptops come with standard M.2 PCIe SSDs that are removable/upgradeable*

*Except for the Samsung Galaxy Book 4, which has soldered UFS storage.

I think you are. Performance and battery life don’t matter when the compatibility isn’t there. Don’t believe me?

From the surface laptop on Amazon:

Did you even read my comment fully before replying? I didn't write that long and detailed comment for the laughs.

If you are calling me shortsighted, then you are calling the Semianalysis authors shortsighted too.

Sure, they will slowly bleed them out. But I am not sure it will be Qualcomm, I can see NVidia in this role. Maybe M$ doing their own chips, or using custom AMD ones. But let's be serious after 8 years of trying Qualcomm was able to produce one decent SoC, that is already obsoleted by its own follow-up mobile SoC, that is overbuilt for the purpose it's serving.

My comment was broadly about ARM, not only Qualcomm.

Well yeah, Qualcomm fumbled the first generation of Snapdragon X/Oryon CPU, but that does not necessarily mean they will repeat the same mistake with their 2nd generation of Snapdragon X. It remains to be seen what will happen.

Where Qualcomm fails, Nvidia and Mediatek might succeed. Nvidia in particular is an established player in the PC industry, and they command incredible respect and influence. You know Windows-on-ARM is getting serious when Nvidia joins the bandwagon.

 

[soresu]

You know Windows-on-ARM is getting serious when Nvidia joins the bandwagon.

Qualcomm might have fumbled WoA in general, but they are still the big daddy of the non Apple ARM world.

The only thing I think that nVidia really has to bring to the table is developer relations and subsidizing software development to accelerate the as yet lackluster native app porting effort.

Prism will never be anything better than a bandaid for compatibility - as all emulation is just wasting compute/power and therefore battery life.

Native ports of major Windows software are needed for WoA to really kick it up a notch.

nVidia have actually done this before when they managed to coax a few game devs to allow access to their engine code to port it to ARM64/Android for SHIELD TV, including Half Life 2.

 

[Gideon]

The only thing I think that nVidia really has to bring to the table is developer relations and subsidizing software development to accelerate the as yet lackluster native app porting effort.

They also have actually working drivers for their GPUs, unlike the still extremely inconsistent support by Qualcomm.

 

[FlameTail]

As Tom Petersen said in this podcast, GPUs are becoming more and more important as a component in SoCs.

If you look at Apple M chips, the percentage of die area taken up by the GPU has significantly increased from M1 to M4.

Qualcomm will have to significantly invest in both software (drivers) and hardware (architecture) aspects of their Adreno GPU. If not they'll completely lose the Big SoC market to rivals like AMD and Nvidia.

 

[soresu]

If you look at Apple M chips, the percentage of die area taken up by the GPU has significantly increased from M1 to M4.

That's just a case of Mx growing into its intended market rather than being just a souped up version of Ax for smartphones and tablets.

The GPU percentage in smartphone/tablet SoCs probably hasn't changed much in a while because that market is pretty mature at this point.

 

[Doug S]

They also have actually working drivers for their GPUs, unlike the still extremely inconsistent support by Qualcomm.

I think that is by far the biggest item in favor of Nvidia. They don't have to beat Qualcomm's Nuvia CPU designs, they just have to reach par with ARM designed cores (or use ARM designed cores if they can't) and they will easily beat out Qualcomm for the ARM PC market. People aren't going to care about another 10% or whatever performance difference on the CPU side when there's a 50% performance difference on the GPU side. Games are also already tested and debugged against Nvidia drivers/GPUs, where it is gonna be a crapshoot with Qualcomm SoCs using their own GPU no games are tested/debugged against.

Nvidia is also a name well known and trusted in PC circles, giving them a further leg up for less engaged consumers (i.e. who don't follow benchmark pissing contests and don't know a Nuvia core from a Via C3 CPU)

 

[jdubs03]

I think that is by far the biggest item in favor of Nvidia. They don't have to beat Qualcomm's Nuvia CPU designs, they just have to reach par with ARM designed cores (or use ARM designed cores if they can't) and they will easily beat out Qualcomm for the ARM PC market. People aren't going to care about another 10% or whatever performance difference on the CPU side when there's a 50% performance difference on the GPU side. Games are also already tested and debugged against Nvidia drivers/GPUs, where it is gonna be a crapshoot with Qualcomm SoCs using their own GPU no games are tested/debugged against.

Nvidia is also a name well known and trusted in PC circles, giving them a further leg up for less engaged consumers (i.e. who don't follow benchmark pissing contests and don't know a Nuvia core from a Via C3 CPU)

To add on. Keep in mind at least in IPC, the X925 is superior. So if the successor version can gain on both the IPC side and on frequency that will remain competitive on the CPU side. With the GPU. It just extends the competitive advantage.

 

[SpudLobby]

I think that is by far the biggest item in favor of Nvidia. They don't have to beat Qualcomm's Nuvia CPU designs, they just have to reach par with ARM designed cores (or use ARM designed cores if they can't) and they will easily beat out Qualcomm for the ARM PC market. People aren't going to care about another 10% or whatever performance difference on the CPU side when there's a 50% performance difference on the GPU side. Games are also already tested and debugged against Nvidia drivers/GPUs, where it is gonna be a crapshoot with Qualcomm SoCs using their own GPU no games are tested/debugged against.

Nvidia is also a name well known and trusted in PC circles, giving them a further leg up for less engaged consumers (i.e. who don't follow benchmark pissing contests and don't know a Nuvia core from a Via C3 CPU)

While I directionally agree and always have, discussed many times here, there’s a caveat.

Oryon V3 might be a bigger jump than the X930 for one, and I think for many the scenario is actually reversed: if Qualcomm ships an Oryon V3 with a 25% performance lead over an X925 at similar power, if not more at still same levels, which is very reasonable given they already have a slight lead and the X925 is what’s coming with Nvidia first, not the X930, then I would take Qualcomm and their improved GPU with Adreno 830/840 for laptops.*

For me among others the GPU is a “good enough” function is a certain bar for games. It also depends on the price tiers too.

Anyway If you’re talking about a 30% faster performance/W GPU for games and a 20-30% faster MT and ST CPU or specifically 20-30% faster (and still less energy) compilation, file compression etc. Then I’d take the faster CPU. I suspect many developers and creatives feel the same, and media encode/decode blocks are not complicated either.

*an if is local AI & GPU/NPU acceleration of creative apps. Depending on just how good Nvidia’s is and for what apps specifically, and how good Qualcomm’s is and continues to be (they have partnered with Adobe and DaVinci so far), could go south for Qualcomm.

The former is still important and will be going forward but also overrated for anything big or really powerful due to memory constraints, and high bus width +datarate LPDDR5 & 6 locally will only change that so much. Doing it in the cloud is where frontier AI will continue to be, and also more efficient computationally due to batched inference.

Anyway I can see Qualcomm having some differentiation that still matters, but given Arm is still good at doing generic grade Apple/Nuvia unlike AMD/Intel, Nvidia should hit the good enough function on CPU efficiency and performance, and for games their SoC will be killer.

It’s possible I’ll change my mind too. Need to see what Qualcomm’s drivers and next GPU looks like, along with pricing, and what NV’a Tegra’s look like.

 

[SpudLobby]

3 Tegra SKUs coming per Dylan Patel.

 

[FlameTail]

Qualcomm Project Glymur: Leaker reveals details about possible new desktop Snapdragon X Elite Gen 2 chipset

Qualcomm's next Snapdragon X Elite series may contain more than just laptop chipsets. Following Qualcomm's CEO claiming that Oryon V3 processors will bring 'excitement back to computing performance', Roland Quandt has detailed early details about a possible desktop-class Snapdragon X Elite 2 SKU.

www.notebookcheck.net

Now, the well-known leaker reports that Qualcomm may have bigger plans for Project Glymur than lightweight laptops. Allegedly, Qualcomm is testing prototypes with an All in One (AiO) 120 mm cooler. Thus, Quandt has inferred that 'Qualcomm is coming for the desktop', rather than just improving upon first-generation Snapdragon X series chipsets.

Well, Glymur has a hulking 18-core CPU block, so that's going to need some substantial cooling.

 

[igor_kavinski]

Things got really serious if QC is going to release desktop machines.

 

[Meteor Late]

While I directionally agree and always have, discussed many times here, there’s a caveat.

Oryon V3 might be a bigger jump than the X930 for one, and I think for many the scenario is actually reversed: if Qualcomm ships an Oryon V3 with a 25% performance lead over an X925 at similar power, if not more at still same levels, which is very reasonable given they already have a slight lead and the X925 is what’s coming with Nvidia first, not the X930, then I would take Qualcomm and their improved GPU with Adreno 830/840 for laptops.*

For me among others the GPU is a “good enough” function is a certain bar for games. It also depends on the price tiers too.

Anyway If you’re talking about a 30% faster performance/W GPU for games and a 20-30% faster MT and ST CPU or specifically 20-30% faster (and still less energy) compilation, file compression etc. Then I’d take the faster CPU. I suspect many developers and creatives feel the same, and media encode/decode blocks are not complicated either.

*an if is local AI & GPU/NPU acceleration of creative apps. Depending on just how good Nvidia’s is and for what apps specifically, and how good Qualcomm’s is and continues to be (they have partnered with Adobe and DaVinci so far), could go south for Qualcomm.

The former is still important and will be going forward but also overrated for anything big or really powerful due to memory constraints, and high bus width +datarate LPDDR5 & 6 locally will only change that so much. Doing it in the cloud is where frontier AI will continue to be, and also more efficient computationally due to batched inference.

Anyway I can see Qualcomm having some differentiation that still matters, but given Arm is still good at doing generic grade Apple/Nuvia unlike AMD/Intel, Nvidia should hit the good enough function on CPU efficiency and performance, and for games their SoC will be killer.

It’s possible I’ll change my mind too. Need to see what Qualcomm’s drivers and next GPU looks like, along with pricing, and what NV’a Tegra’s look like.

ARM standard core is further behind that it may seem at first glance, because, while the performance and efficiency difference is small (in favor of Oryon), there is a decent difference in area, Oryon V2 is quite a bit smaller.
So we have two factors that lead me to believe Qualcomm will be ahead:
-Good difference in area, there is much more potential to be realized by Qualcomm there
-Qualcomm has really good talented Apple engineers with the Nuvia team.

In fact, Apple P core is like 40% larger or so than Oryon V2, depending on which source I've checked. It wouldn't surprise me if Qualcomm catches Apple in single core in not much time.