really hope you are right...but with windows+QC and their surface pro X failed...i keep my exceptions low, but nevertheless i hope you are rightPretty good chance these are close to "real testing". Windows 11 version 24H1 has effectively gone "release ready" by the time they held these 'hands on" labs at Qualcomm. Additionally, there have been complaints from system vendors that Qualcomm is requirement that they use Qualcomm's power management chips. That will lead to less vendor variations from reference designs these tests were run on. The SoCs themselves are going to be the same.
Perhaps if there was bad thermal management in the chassis that could lead to throttling, but given the more thermally challenging AMD/Intel SoCs likely being placed into same chassis ( maybe some largely cosmetic variation) that isn't likely either.
The X Elite are arriving relatively 'late' given the initial , overly optimistic, estimates by Qualcomm. There has been lots of time to spend getting things prepped so that system vendors laptops match the reference ones.
This "Nuvia core" is a server focused core that was retargeted on the fly to a laptop service duty. They also got to skip N3B ( reports indicate that they are using some TSMC N4 variant ). This first iteration is likely to be far more a "Frankenstein" project (mashing together Nuvia CPU cores with Qualcomm GPU/NPU-DSP/Memory/etc. ) than the second iteration.
Would need to see what next iteration improvements are before making any sort of real judgements about "how far behind" they are. Whether they are '3 years behind' or not depends upon how much pipelined, parallel next gen development work they have managed to get done. Right now it is more like an iceberg where most of it is likely out of view.
Qualcomm has Phone SoCs, VR SoCs , etc to align also. Unclear if they are going to pursue almost everything all same time or be more deliberate/prudent in expanding the line up. If they dilute themselves too thinly, then they will likely fall closer to '3 years behind'. Haphazard management seems to be a bigger threat than some technological barrier. There are also lots more 'cats to herd' given can't just kill off dGPUs or Windows feature XYZ unilaterally.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Qualcomm revealed X Elite's benchmark scores
- Thread starter sunny5
- Start date
- Sort by reaction score
really hope you are right...but with windows+QC and their surface pro X failed...i keep my exceptions low, but nevertheless i hope you are right
I'm just waiting to see the real products and not only how they perform, but how much they cost.
If they can sell a ChromeBook for over US$1000, how much will a Windows computer on ARM cost? I suspect it won't be inexpensive, until they cut the number of cores to a minimum.
If they can sell a ChromeBook for over US$1000, how much will a Windows computer on ARM cost? I suspect it won't be inexpensive, until they cut the number of cores to a minimum.
If you are obsessing over the Cinebench Perf/Watt score of the M3 , that is probably mainly for the benefit of iPad Air and iPad Pros than it is necessary for the laptops. They don't "have to" hit that if not deploying to those for Windows/Linux. Kind of missing the forest for a tree there. Their perf/watt is in the 2x zone what AMD/Intel are turning. That's what matters. Not M3 deployments to the iPad Pro.
As for the M3 Pro Perf/Watt versus 23W X Elite, it is respectively 37 versus 41. Where is the X Elite 'failing' on efficiency here? Overall 12 cores for both. Note Qualcomm is on an N4 fab node and M3 Pro is on N3B. Sorry, I don't see the 'fail' here at all.
The X Elite doesn't desperately need E cores to eek out the 'win' level that M3 turns in. AMD isn't using E-cores. ( they have some more efficient cores 4c / 5c that make some trade-offs but it is the same base level design). Intel is pouring on E-core counts like ketchup , but it really doesn't buy them any big win in that chart at all. To compete head-to-head against AMD and Intel , Qualcomm doesn't need the same tradeoffs that the M3 makes.
Consumers are not going to buy the SoCs, they are going to buy complete laptop systems. Apple's $400/TB SSD capacity pricing is going to level the playing field a lot for most consumers versus the fanatical performance on E-cores for laptops. Bulks of consumer Windows laptops are selling at lower than Apple's median prices. The bigger threat to Qualcomm success here is being over arrogant in pricing; not the M3's E-cores.
Apple barely sells the M1 MBA anymore. So yeah. [ IMHO, when the iPad Air M1 disappears , I suspect the M1 MBA won't be far behind in fall out of competition. Selling at Walmart is likely as much 'cleaning up inventories' more so that it is a long term plan. If Walmart sells enough that will mutate into (perhaps around Aug-Sept) a M2 MBA's on sale most of the time. ]
Acer generally has some good 'bang for the buck' laptops. On performance, the X Elite definitely competes with the M1 MBA. (Very unclear how look at charts above and not see that. It is generally competitive with the M3. The M1 is further down in performance. ). Is Acer going to be able to match selling 3 year old, "R&D paid for" hardware on price? Probably not. First, because they are going to have lower priced AMD/Intel stuff to sell which they probably won't want to fratricide. Second, because Qualcomm isn't going to lower the prices that low.
If the M1 MBA makes it to 2025 I would be quite surprised.
Pretty good chance Acer pricing will match up with the M2 MBA with a more respectable SSD capacity attached. So again Performance isn't going to be a big selling point for the M2 system.
Entry level dGPUs in laptops are largely dead. AMD's upcoming iGPUs and Intel's next gen iGPU are going to further wipe them out. Where Qualcomm is in a bigger threat is that they don't keep up with the AMD iGPUs, not the Intel stuff.
The consumers in the Windows space are largely not going to put up with banning Nvidia (and AMD) once in the Max GPU range. The M3 Max is reatlively monstrously sized die. Qualcomm isn't currently trying to compete with that all. And in terms of overall volume in the Windows market it doesn't particularly make much strategic or tactical sense all at. Very similar to the claims that Apple is doomed if they don't cover the "4090 performance". It is mostly misdirection from the really significant tactical and strategic issues.
The number of cores like are not going to be cut "to a minimum". It appears they only have one die. There only so many cores you can 'deactivate' before not particularly covering base recovery costs.
This isn't a "cheapo chromebook" SoC. Qualcomm has phone SoCs they can throw at that if they wanted bother.
Intel is throwing mulitple die sizes at Chromebooks. Qualcomm isn't going to cover the same range with the first iteration of the X Elite. AMD doesn't throw as many die sizes at it, but also don't cover the same range with any given generation.
Pretty good chance looking at MBA like pricing but with better base RAM and SSD capacity starting points. There is more flexibility in the Windows space to 'save' on the screen costs also (as an offset for Qualcomm SoC pricing.)
From what I know, Nuvia/Qualcomm is already working on a cut-down SoC to promote battery life and price.
Most Windows UWP apps can be recompiled with a single click if they aren't using x68-exclusive libraries.
Any app-store UWP apps that were designed with mobile phone compatibility might already have the Windows ARM version in the store already. IIRC, it was one of the defaults added to Visual Studio around the time the Surface X was released.
The hardest part will be games, since they're mostly abandoned once they are released and sales dry up.
Do you even read what I’m posting? What does make you assume I didn’t see the difference between X Elite portable at 23W and X Elite performance at 80W? Why do you assume I have difficulty to see the separating line between portable and performance? I’m also talking about Qualcomm’s own comparison when I say ”they compare”. The graphs I posted are not Qualcomm’s but from Just Josh on Youtube. He’s the one showing all benchmarks. Qualcomm just compared 12-core X Elite with 8-core M2.
My point remains. M3 is not the closest variant to X Elite. M3 has only 8 cores, X Elite 12 cores, as shown in the graphs. X Elite should be compared with M3 Pro which also has 12 cores (the reason I posted these graphs). A 12-core M3 Pro is faster in both the single-core and multi-core test in GB than X Elite at 80W. M3 Pro is also faster in the single-core Cinebench test than X Elite at 80W. Again the point is to compare two chips with the same number of cores at the same wattage. For that Apple doesn’t need M4 to ”rise to the challenge”.
Last edited:
I think this is a little silly to compare at this level. The reality is neither side needs to claim victory to that degree. For instance, no one is avoiding the M3 line because it might be slower or might be faster than the Snapdragon X Elite at some wattage. No, if you are used to the Apple ecosystem, you will continue to use the Apple ecosystem and be more than satisfied with the performance.
On the other hand, if the Snapdragon X Elite is released with OEMs and the battery life holds up (say 15-18 hours or something like that) and is still performant, then Windows users will snatch them up...period. Windows needs this to work, and it is all looking good.
Okay? Again, the only people they really need to convince are Windows users to not bail. When Paul Thurott is using a Macbook Air to see what ARM is all about, that's what they need to stop.
The reality is that we need the real stuff to come out and then we can judge it, right? That's what made the M1 Macbooks so amazing. It was us using them and seeing them live up to the hype of awesome performance, little to no fan noise, and completely awesome battery life. I can't wait for Windows users to get to experience that...if it is true.
What surprises me again and again is the atrocious power scaling of the Elite X. Only 30% higher MC performance in Cinebench at ~4x higher CPU power draw... ugh. With it's 12 performance cores at N4P running full steam Elite X is barely faster than N5 M2 Max with 8 performance and 4 efficiency cores...
I thought from the beginning that Qualcomm's claims of energy efficiency of these cores seem odd, but it's reports like these that really make it clear just how odd they are
M3 Pro is doing even better than you think here. Apple's 12-core M3 Pro is a 6P+6E configuration, and it's holding its own against Qualcomm's chip even though Qualcomm's has 12 performance cores. (QC doesn't have an efficiency core design, at least not in this generation of the chip.)
My best guess...
This CPU core came from Nuvia. Before acquisition, Nuvia was designing server chips. Most Arm server chips so far have been focused on very high core counts per socket, for cloud providers who want to run insane numbers of VMs.
Consider Ampere's new 192-core server chip. It has a power budget of 350W for the entire SoC. Even if you assume none of that goes to overhead like memory controllers and I/O, the power available per core is about 1.8W. That's not a lot, so if you're truly optimizing a core design for this space, you'll end up with something which targets relatively modest clock speeds. They'd want decent perf/MHz, could live with mediocre perf/core (by laptop or desktop standards), and great perf/W.
Then Qualcomm acquired Nuvia, and Nuvia's team had to adapt on the fly to a different market. I doubt they were given enough schedule to design a new core fully optimized for desktop and laptop computing. Outside of whatever changes were possible in the time and/or money allotted, the obvious knob to turn is to run at significantly higher frequencies than it was originally targeted for. Since that always means boosting voltage, that would put them out in the nasty, nonlinear part of the performance / power curve. To me, that's a plausible reason why 80W scaling looks so bad. Even at 23W they're likely pushing it higher than a server core wants to run - they're likely already seeing diminishing returns from adding more power per core there.
I expect Qualcomm to get better in future generations. (Assuming the Nuvia team is happy with how Qualcomm runs the acquired company, that is. The acquisition deal probably included golden handcuff packages for the engineers, but once the time runs out on those, if they're not happy at QC, they'll leave. I don't know whether QC's a good employer or not, or how they treat acquired organizations.)
I don't think that this initial set of ultrabooks with ship with third-party graphics. Maybe via eGPU.
What makes you think about CUDA though? That's a bit of a random change of topic.
"already working on" seems likely is a X Elite "v2.0" die/chip. That likely would not ship this year. Qualcomm has talked about moving the cores to more phone appropriately sized chips. Over time they likely will have more than just one die , but for "version 1.0" that does not make much sense. Trying to take on AMD+Intel entire processor SKU line up all at once is a dubious idea. Versus going for something profitable and develop a good/better relationship with the systems vendors where they 'win' (profit) also.
A SoC aimed more deeply at the "race to the bottom" zone of the Wnidows laptops market is fraught with landmnes. ( AMD/Intel discount competition, lower margins thinning out the other components on system ("cheap by association" hole that AMD is still trying to climb out of), etc. ) That is much more easier to navigate if have a steady set of SKUs that make more money to anchor the core R&D spend.
Qualcomm has previously tossed the 'flagship' phone SoC into the low end "Windows systems" space. Mega battery life has been their sales pitch for their solutions on previous iterations. Or do they put something more deliberate in the middle between 'big phone' SoC and X Elite ( and work on three dies at the same time)? If that "lower price and higher battery life" target falls all the way down to the phone level then it really won't be "a bit lower than a plain Mn" target. If it is more of a "Windows first" focus then could loose 4 CPU cores and some GPU die area allocation to result in smaller die. Smaller die die will hit lower a price if can sell additional millions of them. Pruning off the CPU/GPU cores won't be a huge change in battery life.
Nuvia not equal Nvidia. Not seeing where Nvidia tech was acquired or integrated into any of that discussion.
If this is "well server chips have giant gobs of PCI lanes " premise then the flaw is that when Nuvia was aiming at 32-128 cores the overall die was much bigger. This X Elite is down to 14 and has to share lots of die area space with GPU cores ( which need to be fed data at high rates ) , other mundane PC I/O connections ( Wi-Fi , Bluetooth, standard M.2 SSD ,etc ). All of that much bigger die that has lot of edge space to throw at "gobs of PCI-e controller" is gone.
There are multiple subsystem components that go into composing a SoC. Nuvia's CPU cores are only a narrow subset. The I/O , GPU , NPU/DSP, etc. is very likely mostly coming from what Qualcomm had already laid out if they were to use standard Arm design cores. The server I/O requirements are not being pulled down.
"... Additionally, devices powered by the X Elite will offer 8 PCIe Gen4 lanes apart from those used for the SSD and Wi-Fi, support for MAPI cameras, and potentially multi-day battery life. ..."
Qualcomm revealed X Elite's benchmark scores
From what I know, Nuvia/Qualcomm is already working on a cut-down SoC to promote battery life and price.
forums.macrumors.com
" ... For internal I/O, the SoC offers PCIe 4.0 connectivity for NVMe storage. Elsewhere, the company is using PCIe 3 to supply connectivity to their modem and Wi-Fi solutions. No mention has been made of whether there are any free PCIe lanes for further peripherals. ..."
Qualcomm Previews Snapdragon X Elite SoC: Oryon CPU Starts in Laptops
At least x4 of that aggregate 8 are going to taken by the default M.2 SSD that is in the system. That only leaves x4 PCI-e v4 which isn't much. If wanted dual SSDs there would be zero left. If wanted to put a Wi-Fi 7 card that wanted PCI-e v4 lane would have less than 4 left also. Thunderbolt v5 controller ... again the "extra" x4 gone.
Very similar situation to Apple. Qualcomm needs to get lots of Windows/Linux folks optimizing apps for the iGPU that is always present. Not selling any systems with dGPU will focus attention. In later generations, Qualcomm likely will put additional PCI-e lanes into a bigger version of the die due to system vendor and Windows users base pressure. But across their whole line up enabling Nvidia ... doesn't make much sense at all. ( On even smaller dies may not even get the x8 allocation that X Elite has. ) There are likely quirks and bugs to work out in their own GPU stack that a large volume of new users with new application permutations will surface.
The bulk of Windows laptops sold don't have dGPUs. That fraction is only growing over time.
" ... Qualcomm has not ruled out attaching an external GPU to its Snapdragon X Elite platform. ..."
Tested: Qualcomm's Snapdragon X Elite CPU looks like a legit Intel rival
Qualcomm allowed reporters to supervise a benchmarking session testing the Snapdragon X Elite CPU platform, then compare them to other laptops. It's already among the fastest PC processors.
www.pcworld.com
Qualcomm isn't blocking it , but they probably are not going out of their way to enable it either.
The ports are USB4 which technically the rules allow a host system to skip Thunderbolt 3 if they want to. ( In part why Thunderbolt 4 exists to shift more features to mandatory from 'optional'). How capable Qualcomm's ports are remains to be seen ( If just trying to minimally match the effort AMD is putting in they those might work). If some system vendor wanted to throw in Occulink that could probably work also (via 'extra' PCI-e lanes) . But I suspect this is not a core feature the reference systems set out to test.
Decent chance someone will cobble together some linux drivers primarily meant for some other Arm system with this set up and get something that 'happens to work" when booted on some Linux variant.
If the Windows on Arm core OS supports Thunderbolt 3 subset of USB4 and the hardware enables it, then similar issue would open up a range of PCI-e devices. GPUs would then be whether the drivers happen to fall out of the sky problem. ( I don't see AMD rushing to bend over backwards to port there. Nvidia really has bigger 'fish to fry' also unless the numbers are relatively large. )
Pretty good chance the same mindset that says x86 Windows/Linux has CUDA so Windows on Arm should get it too. Once there is a variety of system vendors, someone will run off headed to some niche for additional differentiation and will get a broader ecosystem fall out than what Apple does with the minimal number of system builds.
The GPU vendors may use version 1 systems with external PCI-e expanders (presuming Thunderbolt works) to do 'test mules' to make drivers that might arrive with version 2 (or 3 ). It is going to be tough for Qualcomm to exclude the dGPU/eGPUs over the longer term if want to subsume larger portions of the x86 Windows/Linux space.
cannot wait for these to ship to end these discussions based on charts...we also got these from Apple and on the gpu side was a chart on their best of the best scenario when they compared the top of the line with 4090 if i remember corectly. So i bet QC is doing the same...at least with Apple we know how everything works after 2-3 years...with these..we have to see how their translations cut the performance..i had enough with the BS surface pro x style devices marketing vs reality
I still think from the history and how companies show their chart, that these will compete on "apple side" with Macbook air and nothing more and maybe even their crappy "native" app will suffer in gpu performance compared to the M3
We will see...comments are just guesses for now
But i have to thank all that put the time to post here
From your guesses how ironic would be that their top of the line will cost over 14" Mbp M3 Max but the performance will be half of that?! People will still say Apple is overpriced?!
I still think from the history and how companies show their chart, that these will compete on "apple side" with Macbook air and nothing more and maybe even their crappy "native" app will suffer in gpu performance compared to the M3
We will see...comments are just guesses for now
But i have to thank all that put the time to post here
From your guesses how ironic would be that their top of the line will cost over 14" Mbp M3 Max but the performance will be half of that?! People will still say Apple is overpriced?!
Last edited:
Agree, but still people can buy it in the 14" size with of course some draw backs
The M3 Max 14" base starts from $3000 36gb ram and 512
Lets see their top of the line Elite where it starts...an well build 14" 15" with at least 32" gb ram and 512, im afraid will start around the same
This is an important point. I don't think these Snapdragon X Elite devices will be cheap. They will be the opposite of cheap. On the other hand, in a few months, there will probably be deals lower than what Apple does, but one will have to be patient.
Substantially more folks buy the 14" with a M3/M4 in it. ( Apple has openly stated that the MBA was best selling and that the old MBP 13" was next best selling. That is now probably being split between MBA 15" and M3 MBP 14" but, both are on the 'plain' Mn SoC).
Qualcomm doesn't have to sell everything to everybody on the first generation. The Max really isn't a target market right now at all. Apple skips the luggable desktop replacement laptops that Windows vendors cover... where is the "Doom and Gloom" there? Apple doesn't "have to" cover everything either.
The real deep competition here for Qualcomm is Intel/AMD, not Apple. Displacing a x86-64 SoC in a Windows box is who the actual customers are (the system vendors ). A substantial portion of the M-series comparisions are just so to get the "Apple hype train" than thinking going to do any real damage to Mac sales. Trying to get around Intel bundling sales tactics and just inertia in the Windows space.
The "top" of the X Elite line up is aimed squarely at Intel; not Apple. Intel sells way more than Apple does. They have pushed the die deep into the diminishing returns zones, but that won't matter if Intel has pushed their even deeper into the same swamp. They are just going to be 'less deep' in the swamp and pitch that as a selling point. That top end version isn't going to be strategic/tactically critical to whether this SoC is a success or not. It is likely just aimed at selling a few more binned dies at incrementally higher margins. That is way more efficient than creating a low volume die that can't hit break-even on cost recovery. ( Max can leverage Apple's unilateral move to just dump x86/dGPUs completely. If a user needs a better than a Mn Pro Macbook then they 'have to' buy a Max. There is no other choice. That doesn't exist in Windows space. There are many dozens of system vendors offering orders of magnitude more product choices if will accept Windows as the base OS. Want is in the middle of the zone and volume sales are going to matter most. ).
They aren't really targeting the Mn Pro much either. They will take their 'cherry pick' wins , but it really isn't the focus bulk target market (probably more business 2D/3D graphics focused than gamers or fringe 'creators' ) . They aren't going to cover AMD's Strix Halo either any time soon either in the early part of 2025.
Windows isn't going to use just one 'silver bullet' from one single vendor to come after the Mac products. Breadth and scope is going to be leveraged. Intel is going to pick their battles ( Users focused on Single threaded) and AMD will pick their own. Qualcomm trying to counter everything they do for the first several iterations would be goofy and unnecessary.
I know some of you are very intrigue by all these new marketing specs and names with new cores, new people but QC is just promoting their business, for us the product itself should talk and our own work on these should talk charts and so on ...but at the end for the consumers is windows on arm, another try, maybe this time is a win
Attachments
Last edited:
33 days until the new ARM based Surface Pro/Laptop reveal, followed by OEMs devices at and around Computex.
I have been waiting for those for years.
It's kinda ironic, that once the Qualcom/WindowsOnARM-exclusivity ends, they finally drop a competitive and attractive product on the market...
I know it's only partially on topic, but is have a few question on the ARM based Windows that I am curious about:
I have been waiting for those for years.
It's kinda ironic, that once the Qualcom/WindowsOnARM-exclusivity ends, they finally drop a competitive and attractive product on the market...
I know it's only partially on topic, but is have a few question on the ARM based Windows that I am curious about:
- Does anybody outside Microsoft know how was the ARM based Windows 10/11 made?
- Did they basically copy the classic x86 code and the started to port different parts to ARM or did they build the ARM version from the ground up as slimmer version - that only adds the legacy stuff, if needed and uses the emulation layer to make it work?
- Will we still get Windows 98 and Co. dialogue pop ups and such on the ARM based version?
I think people have been maybe too obsessed with this exclusivity thing. It's not like any company has attractive ARM IP to begin with.
It's exactly the same OS, just compiled for ARM and with appropriate drivers/kernel tweaks.
I don't think we will really know until we see what is released. There will undoubtedly be OS changes announced as well.