Qualcomm revealed X Elite's benchmark scores

[deconstruct60]

Speaking of Rosetta 2, how did Apple create an emulator like Rosetta 2 that was basically perfect in emulating all x86_

It doesn't emulate all of x86. It is smaller than all of x86.
What have is 64 bit userland level coverage with 32 bit operators. ( but not kernel level in either 32 or 64 bits).
Don't get AVX either. Don't get hardware virtualization instructions. don't get kernel level code.

Microsoft does have a re-compilers to native code. It is just structured different for somewhat diverging objectives. scroll back two posts

 

[boss.king]

These benchmarks mean nothing. What is it like in real world use? How hot does it get? Does it throttle? Does it need a fan blasting to run at peak performance?

If Qualcomm can actually compete with Apple, great, but this is a story I’ve heard before and the final result never quite measured up to the hype.

 

[Xiao_Xi]

Qualcomm on Wednesday shared more technical details of the SoC.

For a reference of what Qualcomm talked about, Anandtech did a live blog.

Qualcomm Snapdragon Tech Summit Live Blog: Compute Spotlight

www.anandtech.com

 

[leman]

Qualcomm on Wednesday shared more technical details of the SoC.

For a reference of what Qualcomm talked about, Anandtech did a live blog.

Qualcomm Snapdragon Tech Summit Live Blog: Compute Spotlight

www.anandtech.com

Not much new information about Orion here. Two interesting factoids:

- The 3.8Ghz clock frequency seems to be the "normal boost" (they say "up to"), 4.3Ghz is an additional boost. I wonder how often that additional boost can be triggered in practice
- The Oryon draws 80 watts in MC Cinebench 2024, which is quite a lot! The score should be around 1200, so 30-40% higher than M2 Pro, which is much more reasonable than what they showed with GB6. But that power cost though...

 

[Bodhitree]

Interesting stuff. It seems like the single-threaded performance vs watts curve of the Oryon is not that different from the M2, they just clocked it higher and so are using more power. Using 80W in Cinebench is not a good sign for laptop use, where this extra power draw will come at the cost of extra weight in thermal design and crucially reduced battery life.

Laptops are the key market for this kind of lower-power, high performance cpu design, and as I see it sales will be dependent on offering professional users the magic of great cpu speed coupled with great battery life. It all comes down to proving that marketing promise, of a noticeably better balance than in an x86 laptop which will not be able to use the top-end desktop cpus. And that might depend on how they downclock the Oryon cpu in practice to reach more sustainable power levels for a laptop.

Price is the other important factor. Windows On ARM devices are still quite a young and unproven market, and it seems to me that in order to grow that market Quallcomm would be better off not charging a high premium for their new processor. But we will have to see.

Interesting that they haven’t focussed on multicore benchmarking. With all P-cores and a lot of them, as you’d expect from an originally server-based design, this could lead to a monster multicore score.

 

[leman]

Interesting that they haven’t focussed on multicore benchmarking. With all P-cores and a lot of them, as you’d expect from an originally server-based design, this could lead to a monster multicore score.

They showed plenty multicore benchmarks. In fact, they showed more multicore results than single-core results.

 

[camotwen]

And yet here you are, pointing it out in an Apple centric forum.

Well, a lot of people here have actually complained many times that they cannot dual book macos and windows in the new macs, so this observation is nothing new to an Apple centric forum.

 

[camotwen]

So qualcomm has nice efficiency plots for both intel and amd comparisons, but not for m2. Sounds fishy to me - though I guess anybody can take the numbers and make an educated guess of what that would look like.

But in any case, it is also understandable. Even if they do not beat apple silicon, it is more than enough to beat intel and amd, as anyway they go more for their market share than apple's.

 

[Xiao_Xi]

qualcomm has nice efficiency plots for both intel and amd comparisons, but not for m2.

Are such graphs even possible with Apple SoCs? Other than Apple's marketing material, I haven't seen performance vs. power consumption graphs with Apple SoCs, not even from independent reviewers.

@leman Is it much harder to get the data for the graph with Apple SoCs?

 

[leman]

Are such graphs even possible with Apple SoCs? Other than Apple's marketing material, I haven't seen performance vs. power consumption graphs with Apple SoCs, not even from independent reviewers.

@leman Is it much harder to get the data for the graph with Apple SoCs?

I did publish the graphs, at least what I could make

https://mr-mobster.github.io/AppleSiliconPowerTest/analysis.html

I do not know how to control the frequency of the Apple CPUs, but we can sample the performance counters of individual threads under various circumstances (e.g. different power modes and CPU utilisation), which gives us enough points to glimpse the trend.

 

[MRMSFC]

So qualcomm has nice efficiency plots for both intel and amd comparisons, but not for m2. Sounds fishy to me - though I guess anybody can take the numbers and make an educated guess of what that would look like.

But in any case, it is also understandable. Even if they do not beat apple silicon, it is more than enough to beat intel and amd, as anyway they go more for their market share than apple's.

Yeah, I was thinking it’s far more important for them to show up Intel/AMD considering that’s their direct competitors.

The real interesting part in on-device AI to me. I’m sensing a new area of competition in performance for NPUs.

 

[Xiao_Xi]

The real interesting part in on-device AI to me. I’m sensing a new area of competition in performance for NPUs.

If more and more applications add AI-based features, it is possible that the user experience will depend more on NPU performance than CPU performance.

And if it is true that Qualcomm has a better NPU than AMD and Intel, Qualcomm, with its higher-performance NPU, could help developers deliver more AI-based features.

By the way, Chips and Cheese published an in-depth analysis of Qualcomm's NPU last month.

Qualcomm’s Hexagon DSP, and now, NPU

Cell phones have to handle telecommunications and audiovisual processing while maximizing battery life. Digital Signal Processors, or DSPs, use specialized hardware to offload these tasks from the …

chipsandcheese.com

 

[deconstruct60]

So qualcomm has nice efficiency plots for both intel and amd comparisons, but not for m2. Sounds fishy to me - though I guess anybody can take the numbers and make an educated guess of what that would look like.

Who are Qualcomm's actual customers? System Vendors. How many Elite X SoCs is Apple going to buy? Zero.
So are they a customer? No. Are they even a rational possible customer? no.

"Fishy" would be some sales pitch to sell SoCs to people who are never going to buy it. "I've got this great sell ice to Eskimos in December-January pitch ... want to hear it". errr not really. Is Apple going to sell M-series to the Windows guys ? Nope.

The other part that is not fishy at all is that both Intel and AMD iGPUs are about to take a substantive step function increase in performance in 2024. To even be a creditable option in mid 2024, Elite X has to beat the 2021-2023 options. The other sales pitches these same system vendors are hearing from AMD and Intel as also talking about how much better their iGPUs are. If Qualcomm doesn't present substantive improvements here , they may not even get invited to a 'bake off' to be a contender.

the AMD/Intel gaps Qualcomm presented here are going blackside by the time Qualcomm ships.

The other major shift that is happening in Windows laptop market is a shift to fewer dGPUs systems. It isn't about king of the mountain GPU output. If folks want to get into a tech spec porn war on that iGPUs are going to loose anyway. Apple mandated them all gone. For the system vendors it is really not going to work from a single mandate from top-down to do that.

At these tech summit presentations Qualcomm gave HP , Lenovo, Honor, Microsoft , Asus , etc GOBS of stage time. In both the key and the tech session that covered the Elite X. There is a 'herd' factor that Qualcomm is trying to capture here. If several of the major Windows System vendors have systems with Elite X in them then some other vendors may be some FOMO (fear of mission out) and sign up also. If Qualcomm can't displace AMD/Intel on some significant wins then there will be no FOMO factor at all. AMD and Intel iGPUs are their primarily competitors. Period.


However, none of these system vendor are going to 100% to Qualcomm only. ( nobody is going to do what Apple did. Apple Silicon doesn't have to complete for macOS system adoption. ) . The best case outcome is for Qualcomm to get some deploy to a subset of systems in the Windows laptop/All-in-one market where they are a good fit. Vast majority of folks do not buy macOS systems. Worry about selling to some of those folks first.

So far nothing presented looks like Qualcomm is going after systems that are focused on dGPUs being present. ( not trying to entirely displace Nvidia ) . That is a far more prudent approach. They don't have to display 'everything for everybody' on the first generation. They have probably thrown more at CPU and NPU than a 'main focus primarily on GPU ' approach would do (i.e., Apple 'Max' dies. Which are pretty close to a GPU with other stuff wrapped around it. ).

For windows laptops where battery life has a top ranked priority a 'more than decent' iGPU will get traction on dGPU systems on bill of materials and power consumption. It is going to be attractive to Windows System builders. Qualcomm's best hope is that the savings on power consumption from their CPU/NPU/mini-NPU/etc are also addition on top of what the iGPU brings so there is a substantive aggregate system level impact. The iGPU doesn't have to be the absolute king of the mountain at any cost. It can't be horrible , but 'competitive enough' is going to suffice to contribute to the more holistic overall win.

[ Pretty likely AMD is going to win the 'max performance' iGPU in 2024 , but not at the same power levels. ]

 

[deconstruct60]

If more and more applications add AI-based features, it is possible that the user experience will depend more on NPU performance than CPU performance.

NPU inference locality is more an issue than absolute maximum performance. that factor is getting lots more traction now relative to the mania 3-5 years of ago pushing all AI inference into the far remote cloud at mass aggregation data centers.

Not about the largest possible model but the large enough to be useful (without giving away the farm) model.

Apple somewhat started on the 'more local' path before many , but may have seriously underestimated how quickly competitors could adapt and take that even further and faster.

 

[Bodhitree]

Interesting article to add to this discussion…

Nvidia and AMD plan to launch Arm PC chips as soon as 2025, Reuters reports

Microsoft might be looking beyond just Qualcomm to challenge Apple.

www.theverge.com

If true these chips may be expected to be weak on cpu and strong on graphics. A combination that game developers may find it easier to work around than the opposite.

 

[iAppleOrchard]

This may be true for only 4 more days.

 

[camotwen]

Who are Qualcomm's actual customers? System Vendors. How many Elite X SoCs is Apple going to buy? Zero.
So are they a customer? No. Are they even a rational possible customer? no.

"Fishy" would be some sales pitch to sell SoCs to people who are never going to buy it. "I've got this great sell ice to Eskimos in December-January pitch ... want to hear it". errr not really. Is Apple going to sell M-series to the Windows guys ? Nope.

The other part that is not fishy at all is that both Intel and AMD iGPUs are about to take a substantive step function increase in performance in 2024. To even be a creditable option in mid 2024, Elite X has to beat the 2021-2023 options. The other sales pitches these same system vendors are hearing from AMD and Intel as also talking about how much better their iGPUs are. If Qualcomm doesn't present substantive improvements here , they may not even get invited to a 'bake off' to be a contender.

the AMD/Intel gaps Qualcomm presented here are going blackside by the time Qualcomm ships.

The other major shift that is happening in Windows laptop market is a shift to fewer dGPUs systems. It isn't about king of the mountain GPU output. If folks want to get into a tech spec porn war on that iGPUs are going to loose anyway. Apple mandated them all gone. For the system vendors it is really not going to work from a single mandate from top-down to do that.

At these tech summit presentations Qualcomm gave HP , Lenovo, Honor, Microsoft , Asus , etc GOBS of stage time. In both the key and the tech session that covered the Elite X. There is a 'herd' factor that Qualcomm is trying to capture here. If several of the major Windows System vendors have systems with Elite X in them then some other vendors may be some FOMO (fear of mission out) and sign up also. If Qualcomm can't displace AMD/Intel on some significant wins then there will be no FOMO factor at all. AMD and Intel iGPUs are their primarily competitors. Period.


However, none of these system vendor are going to 100% to Qualcomm only. ( nobody is going to do what Apple did. Apple Silicon doesn't have to complete for macOS system adoption. ) . The best case outcome is for Qualcomm to get some deploy to a subset of systems in the Windows laptop/All-in-one market where they are a good fit. Vast majority of folks do not buy macOS systems. Worry about selling to some of those folks first.

So far nothing presented looks like Qualcomm is going after systems that are focused on dGPUs being present. ( not trying to entirely displace Nvidia ) . That is a far more prudent approach. They don't have to display 'everything for everybody' on the first generation. They have probably thrown more at CPU and NPU than a 'main focus primarily on GPU ' approach would do (i.e., Apple 'Max' dies. Which are pretty close to a GPU with other stuff wrapped around it. ).

For windows laptops where battery life has a top ranked priority a 'more than decent' iGPU will get traction on dGPU systems on bill of materials and power consumption. It is going to be attractive to Windows System builders. Qualcomm's best hope is that the savings on power consumption from their CPU/NPU/mini-NPU/etc are also addition on top of what the iGPU brings so there is a substantive aggregate system level impact. The iGPU doesn't have to be the absolute king of the mountain at any cost. It can't be horrible , but 'competitive enough' is going to suffice to contribute to the more holistic overall win.

I literally said the same in the second part of my post that you did not quote. But I do not care actually; I am gonna buy the best system available. If qualcomm beats apple, I am gonna buy qualcomm and use Linux. Linux on arm is actually a quite advanced system, which has plenty of software I care about actually working fine there already. And its prevalence will increase. This is why I want honest reviews and tests, to be able to get the most informed decision. And we do not see these from any manufacturer ever, actually. So yeah, when I see Qualcomm obscuring things with the apple comparison, it smells fishy to me, as in the want to present only positive things about their product, so they hide anything that may not be as positive. I doubt qualcomm will get even close right now.

Moreover, nobody expects that qualcomm will just replace intel and amd completely. First generation of M1s went side by side with the last generation of intel macs, for the unfortunate souls that made the mistake to get those. Moreover, people who for some mysterious reasons insist on gaming on heavy, overpriced laptops with power hungry gpus which sound like jet engines and last 20 minutes on battery will continue to exist, and for those whatever iGPU qualcomm will be using will not be satisfactory, also as will the need for fast professional workstations for now for which electricity bills and fan noise does not matter as much. And there are a lot of unknown with igpu etc (reminder that apple silicon has the best iGPUs right now by far, and they actually compete to some aspects with dGPUs really well).

However, I think that the software world is finally ready to get -ast the x86 era, and that companies cannot get away with avoiding native ARM support anymore, even in windows.

 

[MRMSFC]

View attachment 2302537
This may be true for only 4 more days.

Maybe, but you race with the car you brought.

 

[Juraj22]

Maybe, but you race with the car you brought.

That is correct. And SnapdragonX laptop you can actually buy will be available in 2024. So it will compete with M3 class cpus from apple that you might be able to buy in 2023. This feels like Qualcomm had to make announcement because otherwise they would need to remove comparison with M2 and redo it for M3 were it might not be that rosy. Anyway, it is good that competition started. Intel & AMD are not exactly producing right HW for laptops. Yeah, AMD is a bit better..but still..

When Qualcomm cpu in laptop can report something like this:
CPU Power: 170 mW

GPU Power: 25 mW

ANE Power: 0 mW

Combined Power (CPU + GPU + ANE): 195 mW

many users will be happy. This is typical consumption for task "browsing the web" on M1 Max

 

[Kpjoslee]

Pretty exciting stuff. I doubt Apple even bothered to reveal M3 this year if it wasn’t for Snapdragon X announcement.

 

[thenewperson]

Pretty exciting stuff. I doubt Apple even bothered to reveal M3 this year if it wasn’t for Snapdragon X announcement.

lol, that definitely wouldn’t have any bearing on the release of their chips. Lead times are just too much for that to be possible.

 

[vanc]

That is correct. And SnapdragonX laptop you can actually buy will be available in 2024. So it will compete with M3 class cpus from apple that you might be able to buy in 2023. This feels like Qualcomm had to make announcement because otherwise they would need to remove comparison with M2 and redo it for M3 were it might not be that rosy. Anyway, it is good that competition started. Intel & AMD are not exactly producing right HW for laptops. Yeah, AMD is a bit better..but still..

When Qualcomm cpu in laptop can report something like this:
CPU Power: 170 mW

GPU Power: 25 mW

ANE Power: 0 mW

Combined Power (CPU + GPU + ANE): 195 mW

many users will be happy. This is typical consumption for task "browsing the web" on M1 Max

That's a joke, right.

It's easy to measure the power consumption in real-time:
```
sudo powermetrics -i 1000 --samplers cpu_power,gpu_power -a --hide-cpu-duty-cycle --show-usage-summary --show-extra-power-info
```

On my MBP 16" with M1 Pro, combined power was around 1w to 3w when playing 4k Youtube video. If you load/refresh a web page, peak power usage could shoot to 5w and more depending on the complexity of web pages.

Apple's official spec for 2021 M1 MBP 16" was 14 hours wireless web. Given the 100Wh battery, it ends up with about 7.14w power draw for the full system (including the LCD panel with 50% brightness (8 clicks from bottom)). That seems fair.

But again, the power consumption of Apple Silicon is amazing.

 

[APCX]

Pretty exciting stuff. I doubt Apple even bothered to reveal M3 this year if it wasn’t for Snapdragon X announcement.

Yes, I hear they were thinking about taking the year off. Surely Tim Apple would approve?

 

[XboxEvolved]

Hold on, you’re telling me a CPU that isn’t out yet, that is coming out in the middle of 2024 is more powerful than a processor that came out in January 2023? That’s crazy! Something a year and half old being more powerful than the older thing has never happened before! What ever will Apple do, since a processor that will be a year and half old by then will get smashed by something that isn’t even in production yet? Who knows? Oh wait…

 

[itsboi]

Hold on, you’re telling me a CPU that isn’t out yet, that is coming out in the middle of 2024 is more powerful than a processor that came out in January 2023? That’s crazy! Something a year and half old being more powerful than the older thing has never happened before! What ever will Apple do, since a processor that will be a year and half old by then will get smashed by something that isn’t even in production yet? Who knows? Oh wait…

It's even more crazy when the M2 Pro/Max series were supposed to be late 2022 releases.