Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Snapdragon X Elite Multi-Core Superiority Claims over Apple's M3 Were Right (wccftech.com)
54 points by rbanffy on March 16, 2024 | hide | past | favorite | 62 comments


This part in the article feels a bit sloppy: "A fair comparison would be pitting the Snapdragon X Elite against the M3 Pro, but it is likely that the more capable Apple Silicon would obtain a higher multi-core score."

OK, so to get the scores for M3 Pro, go to GeekBench Mac benchmarks (https://browser.geekbench.com/mac-benchmarks):

Scores: M3 Pro (12 core): 15274 M3 Max (16 core): 20958

Vs. Elite (12 core): 13925

Still impressive though.


I can’t tell if Snapdragon’s Elite branding is supposed to be the same as Apple’s Pro. If that’s the case, then comparing it to a base M3 is not appropriate.


Form factor and pricing will play a big role. You can’t get the M3 Pro in a MacBook Air, and it costs a bit more than the base MacBook Pro models.


> In multi-threaded workloads, Geekbench 6 reveals that the Snapdragon X Elite is actually 16 percent faster than the M3 and trails behind the latter in the single-core run.

What's impressive to me is that Snapdragon X Elite is already competitive despite being built in an older N4P tech (announced in 2021). IIRC Apple's latest offerings are built using N3B.

So Qualcomm is onto something here.


> So Qualcomm is onto something here.

Yeah. Using 12 cores to get 16% more performance than an 8-core processor gets.


The odd thing is that the Qualcomm CPU is 89% as fast as the M3 in a single core test.

It shouldn't take 12 cores to achieve dominance.


Power limits probably affect this. And scaling cores isn't linear, their cores clearly scale poorly.

Though Geekbench particularly scales poorly with cores.


Yeah, that would be like using 12 valves in an engine instead of 8 valves to get 16% more performance!



I think pistons instead of valves would work better for your analogy.


I thought about that for a hot minute before making the analogy and in previous years I would have agreed. The cores are doing the work, after all. But CPUs nowadays are so different nowadays compared to yesteryear.

In one CPU you have cores of different sizes and different speeds, and there are various ways to connect them to each other, and interplay with cache subsystems etc.

Anyways it doesn't matter. If I put in a certain number of watts and money and get a certain performance out, what does it matter how?

(It's not like I when I had to buy 8 sparkplugs for my V8 engine instead of 4 sparkplugs for my straigt 4 engine when doing service.)


> In one CPU you have cores of different sizes and different speeds

In this case yeah, but not all of them. I believe AMD still makes some desktop CPUs without those compromises, but they're still slowly starting to put out CPUs like the 7950X3D that have crippled cache on certain cores (7950X does not, 7800X3D does not).

Of course big.LITTLE has been a thing in ARM for decades, including ASi. But I believe Intel started doing E-cores because they were starting to run into space and thermal constraints, and now AMD is exploring heterogeneous architectures as well. I hope they won't repeat the same mistakes.


They are certainly different cores.

But above a certain threshold of performance, what matters to me as a consumer is perf/$.

Also keep in mind that they compared to the MacBook Air model which doesn't have fans for cooling and Geekbench is known to be a short burst benchmark. So Qualcomm's multicore performance lead might be larger for continuous usage.


The Qualcomm has 12 high performance cores. The Apple one has 4 performance cores and 4 efficiency cores that deliver 1/3 to 1/2 the performance. The Apple M3 Max had 12 performance cores and 4 efficiency cores, but that’s also a lot more expensive.


I mean...............yes, you're right. But at the end of the day if the CPU is faster at the same energy used(I assume it is), then the end user won't care whether the CPU is 1, 8 or 128 cores(and why should they).

It's like people saying that Apple products are inferior because they usually have less ram than competition - and yes, that's technically true, but in practice it literally doesn't matter, they are still usually just as fast if not faster than competition with more ram - so why should the end user care.


That's just stupid wishful thinking from all the Apple propaganda. With the limited amount of RAM they have, the base level Macs are not as able as some other laptops that may be slower in computational power, (at least on paper and in short Geekbench). In fact, the moment you end up using swap, some workflows can become twice as fast on much slower and way cheaper laptops.

There are countless videos/ressources illustrating that by now so you may want to stop talking shit.

Some people may be just fine with 8GB but for most peoples who actually want to pay a bit more for computers this limitation will come up pretty fast, in fact I'm pretty sure many peoples heavily using computer have a baseline usage of around 8Go bare minimum (OS plus regular utilities plus minimum web browser load).


It’s not the same energy used, the Snapdragon uses more


And without unified memory I assume, and probably smaller caches.


What I really wonder is if these new Snapdragon SoCs will be closer to what we're used to with PC hardware today. e.g.: Will we get SystemReady and the ability to disable Secure Boot?

There doesn't seem to be any information regarding that, so I guess probably not. But after a lot of false starts, it does seem like competitive alternatives to x86 computers that are also more "open" PC platforms are probably happening in the imminent future. This would be nice.

I suspect that what isn't imminent is seeing stuff like this show up in desktop parts, but if it does, here's to hoping for modular main boards with DIMM slots.


You can disable Secure Boot on at least some of the existing Snapdragon-based ARM PCs already.


All. Every single arm64 desktop machine supports disabling UEFI Secure Boot.


However is there support cycle less than x86? Not only that there's a lot more fragmentation. Is there backward compatibility to run aprogram 6 years ago for an arm hardware that was released this year?


I might be missing some examples but the last breaking change i remember was the deprecation of the 32bit support: https://www.gsmarena.com/arm_will_drop_32bit_support_in_the_...

I think ARM are clearly more comfortable breaking compatibility than Intel.


Wouldn't performance per watt be the correct measurement for this class of device ?


Correct. Everything QCOM has been demoing has been at two power points: 23W and 80W. These tests look remarkable but the real test's going to be out in the field when they're power-throttled to nothing and back in hardware designs trying to cut every cost they can, including their power budget.

It's great they made a Hemi engine, but it's not going to fit in the Smart car that companies want to put it in.


Or Performance per Dollar?


Performance per dollar has traditionally been the selling point with PCs. Performance per watt is the selling point with Apple silicon.

Does this one fit somewhere in between?


ASi has performance-per-dollar for days if you need the unified memory. i.e. it's more performant than some datacenter-class GPUs while potentially also having far more memory as well.


This is such a bizarre article. First, they say superiority claims were right all along, and then paragraph after paragraph they reveal worse single core performance, worse multi core performance (needs 4 more cores to gain 16% advantage), higher power draw, and to finish it off they mention that Geekbench is not really a great benchmark.


Not only 4 more cores but the QC chip has 12 performance cores rather than the 4+4 setup of the M3.

So 50% in total core advantage and 200% when it comes to high performance cores only for 16% multi-threaded lead is hardly an advantage.

Ofc there are caveats to this if QC’s performance cores are closer to Apple’s efficiency cores in terms of size and power consumption then it’s a different story. But that doesn’t seem to be the case.

The other major caveat is that QC has so far only been demoing its SoCs on devboards which don’t share the same power and thermal restrictions as real world designs. Whilst the M3 figures are for real world hardware.

So squeezing the Snapdragon Elite into a MacBook Air style form factor might paint a very different picture.


Apple performance cores are very big in terms of area, you can't just compare the numbers.

I think it comes down to performance / watt / area, while single core performance is also important.


What's the software story expected to be like with these desktop Qualcomm chips? Any chance of official Linux support, or are we just getting Windows drivers and maybe semi-working reverse engineered Linux support years down the line?


Thinkpad has a ARM Snapdragon model and people have been working on it in Linux for a while. Linux support is unofficial, and still crap, years after release.

For example:

https://news.ycombinator.com/item?id=37431710 ("Linux on the Arm-Based ThinkPad X13S: Getting There")


System76 hopefully listening!

(To prevent rumors, I’ve never seen them mention arm support before… maybe someday)


I don't think there's much System76 can do if the ARM SOC vendors aren't willing to cooperate by openly releasing code and documentation for their parts.



> But still a work-in-progress is audio support, DP Alt-Mode, enabling the DSPs, USB-C power delivery, and GPU acceleration.

AKA still garbage.


Usually vendors of ARM64 chips provide the Linux support straight from the off. What's the story here that Qualcomm doesn't?


Rumors are of mainline support pretty quickly.

But the proof will be when it actually happens.


Hopefully it gets chrome OS support so that it has official Linux drivers.


I wonder why the snapdragon does only slightly increase in compute power even tho they nearly triple the power consumption, or am I reading the charts in the source (linked on page) [0] wrong?

https://wccftech.com/qualcomm-unveils-more-snapdragon-x-elit...


If there is an always-on baseband processor that loads a binary blob or like the Pi, where the GPU actually manages the initialization and running the ARM CPU, then it's a no from me... without open source bootloaders we'll still be subject to the tender mercies of Qualcomm.


- 12 cores vs 8

- 8 perf. cores vs 4

- higher TDP

- ~$1,968 vs $1,499

not a fair comparison, you need to compare against the M3 Pro, and then you loose ;)


I would have been shocked if Snapdragon had suddenly beaten Apple silicon because it would necessitate such a huge increase in performance, that it's nearly impossible.


It looks like the future for laptops is ARM. I hope Lenovo makes a ThinkPad with this chip, there already is a ThinkPad with a Snapdragon SoC (ThinkPad X13s), but from what I have read it's not very good. I considered getting a MacBook to use with Asahi but some important features are still unsupported (like HDMI output support).


Does Windows run natively on ARM + most of its software?

Do they have their own translation layers as fast as Rosetta 2?

Just curious as to how it works.


Yes, Windows 11 runs natively on ARM. There is ARM native versions of MS Edge, Office, Visual Studio. Most software from 3rd parties is still x86/x64 based but some things like Firefox are available for ARM.

They do have their own translation layers for both x86 and x64 based apps. It's "slower" than Rosetta 2 but it's just commodity ARM hardware with no accommodations for certain x64 instructions like Apple Silicon. I've been mostly impressed by it.


Oh, for a second I thought you meant the limited VS Code, not the real Visual Studio - but I checked and they do actually have a proper full VS running on ARM now, that's super cool. Would make me seriously consider buying an ARM windows laptop.


Rosetta 2 was/is "impressive" because Apple was smart enough to bake some x86 functionalities into their ARM designs to aid the transition. On the other hand Qualcomm and other OEMs don't have in their interests to offer such functionalities so Microsoft is forced to 100% emulate x86


Last time I dug into this, the conclusion seemed to be that Windows' translation layer, built in cooperation with Qualcom themselves, is actually quite competitive. Performance numbers weren't that far off from Rosetta for desktop applications, with an estimated performance 0-15% lower than Rosetta, which is quite low for a platform without x64 memory model optimisations.

The first release could only do one architecture (32 bit or 64 bit, I'm not sure) but its capabilities have been expanded. A lot of benchmarks are outdated and it seems like nobody has been paying much attention to it in a while now.

The biggest problem with Windows on ARM doesn't seem to he the binary translation, but the slow SoCs the translation is running on.

The most important tools (browsers, Office) work fine on ARM without translation. The translation layer is necessary, but I'm not sure if common use cases will see all that much use for most people.


It’s definitely performant. I’ve been using it for mofe than a year under parallels on an M1pro macbook to run solidworks. It runs solidworks about as well as my 2019 13” intel MacBook running bootcamp. It runs non-GPU intensive stuff much better.


I don't know exactly what model you had, but the truth is 13" models from that time were a complete joke in terms of power for the price.

Beating this in a thicker chassis with a more efficient processor is not exactly an achievement. Non optioned 13" models still suck in terms of power for money, but at least in the Apple Silicon age they are less ridiculous.

But if you don't fanboy and don't need the battery life you should realize they are still are very bad deal in term of capabilities.


On windows for ARM there are quite a few bugs during emulation that I don’t get on an x64 based system, even while using effectively the same settings for the program. Software using custom installers will be the biggest issue, and this is significantly more common on windows than on macOS


Yes, windows/arm64 runs all arm64 binaries out the box. Looking at my jenkins windows/arm64 CI machine, i'm running x64 java, but that's it.

If you want to build apps to run with support for x64, there is a special Arm64CE target type which changes how binaries are built to make compatibility with x64 better (as in faster, lower overhead).

It's described here: https://learn.microsoft.com/en-us/windows/arm/arm64ec-abi


Perhaps I'm out of date on the subject, but does it matter how fast it is when Apple's x86 translation layer makes Microsoft's look like banging rocks together? Not to mention the fact much of Apple's ecosystem now runs natively on ARM whereas that is decades away for Microsoft if it ever happens at all.


A native app list: https://armrepo.ver.lt/


I wonder what type of applications we use daily makes use of multiple cores?


Web browsers typically run separate tabs or groups of tabs on separate cores to spread the workload more evenly.


Operating systems with lots of background tasks and threads.


When can we get a device with it though?


"However, bear in mind that the Snapdragon X Elite was previously tested at two power limits; 23W and 80W, so it is possible that the SoC obtains a higher single-core and multi-core score with the bigger power draw at the expense of heat and rising temperatures, not to mention battery life if the Galaxy Book 4 Edge would be used without being plugged in."

Meaning this comparison is relatively pointless at this stage until third party unbiased apples to apples comparisons in controlled environments are possible. Comparing Geekbench scores of different CPUs operating at unknown power limits in unknown environments are pointless.


The initial results are still looking impressive though. There is finally good non-Apple ARM.

Time to move the goal posts and see if they can actually get Windows or Linux working as well as MacOS on ARM.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: