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
If these will act and perform as the M1 and M1 Pro will still be a step forward
Yes but if they perform like a celeron then it may cement the PC OEM world into x86. Or at least convince consumers that ARM just isn’t worth it.
Exactly, which I think would be a shame. The old-style x86 is a power-hungry dinosaur, and it would be a good thing for the planet if more power-efficient architectures took over. So from that angle I was kinda rooting for the Snapdragon X1 Elite chips, thinking that they might prove that a credible alternative to x86 exists.
Although if the Snapdragon X1 Elite fails, that’s not the end of the line for more power efficient architecture. Intel’s Core Ultra series look somewhat promising too, since they mix low-power-consumption x86 cores with high-power ones.
According to Android Authority, it seems so.
X1 Elite (X1E84100) X1 Elite (X1E80100) X1 Plus (X1P64100) Total package power (95% parts) 98.50W 52.92W 42.52W Total package power (50% parts) 82.33W 43.40W 42.52W
Exclusive: Here's what Qualcomm didn't tell you about the Snapdragon X series
The upcoming Snapdragon X1 will be available in five different models: see their specs and performance here.www.androidauthority.com
Context matters. Right above that table
"...
In practice, the better parts will simply get binned as the higher-end SKUs, though, so the distinction doesn’t have any significant implications in reality. However, it’s still important to keep in mind.
| X1 Elite (X1E84100) | X1 Elite (X1E80100) | X1 Plus (X1P64100) | |
|---|---|---|---|
| Total package power (95% parts) | 98.50W | 52.92W | 42.52W |
| Total package power (50% parts) | 82.33W | 43.40W | 35.01W |
These pre-binned-and-tagged numbers ( xx% parts ). The X1E-84-100's should come out of the 50% parts pile ( half of the parts meet this binning. If there is high demand for these , they can can keep up with their supply). Pretty good chance the X1E-80-100 will also (unless demand is far too high. e.g., 84 class priced to low. ). X1P-64-100 will mostly come out of the set difference of between 95% pile and 50% pile. In short, about half of those Wattage numbers probably shouldn't show up in any end user system. The chart more so is about how easily they can fill the categories.
Pretty good chance the '78' can come from the 95% pile also (no turbo 2 core and same clocks as the 64.). If the 84 and 80 don't consume all of the 50% pile, nobody will complain much with 70 on down coming from that pile.
that 50-50-95 sorting would put them in the 3.6-7.8 W per core range; probably mainly shooting at about the 42.5W target ( sometimes over and fewer sometimes under). Versus AMD and Intel do they really generally have a big problem? The '84' class versus the 75-85+W AMD/Intel parts probably won't be a 'huge hit', but it probably doesn't "have to be" a 'huge hit' either. If the 85 is too soft Qualcomm can throw a Plus model out there that will shoot for the 35W zone to soak up the excess. ( for now they'll just wait; 85's will make more money if they sell enough of them. )
There are several Android focused sites (like Android Authority) selling the notion of P and E cores. I think that is a confusion about the benchmarks reporting two processor clusters and the desire to throw the different cluster into separate categories. It won't be surprising if the core cluster thing is more so to support the ( 2 core Turbo ) thing more so than any P versus E core like difference (i.e., running the clusters at different speeds like 'on' and all the way 'off' ). When the two 'golden sample' 2-core-Turbo cores end up in one cluster, they just turn off the other cluster when running in that mode. ( not really P vs E as much as scaling unused resources back so they can run 'Turbo' longer. ). "Turbo" drag racing is not the primary focus of these cores. Lack of hyper single threading drag racings doesn't necessarily make it an 'E core'. There is likely just one CPU core design here laid out just one way.
Consumers making gross observations on "as simple as possible" metrics like 'core count' is one reason why it isn't the same as the M3 ( actually 8 ) . Intel throwing 'lower than P cores" at their offerings is boosting the core counts versus AMD. If AMD has '10' or '16' than Intel wants to throw '12' and '18' or '24' counts at the consumer. Nuvia was initially in the server business and would have need >64 , if not > 100 , cores to be a player. It is doubtful they were ever aiming at a number like 8-10. That 'core count' inflation factor in the competition with AMD/Intel always meant that the base count here was not going to be the plain Mn SoC.
This 10 core version is just a binned down version of the 12 core die. They are marketing it with a different label slapped on top, but it is really pretty much the same thing; just might have a 'defect' or 'quirk' that they can skip around.
The '8' and '10' core variation that the M3 has is the GPU cores; not the CPU cores. Once have to account for CPU and GPU core counts then moving past the "as simple as possible" consumer comparison metrics. Thrown in the AI/ML cores and that will be up to 3 dimensions of measurement. GPUs are even more likely to be coupled to Intel iGPU metrics than the CPU cores. Qualcomm likely isn't out to 'out compete' with the Mn Pro iGPU at all. ( nor the top end range of AMD. ) . For one thing, I doubt Qualcomm is trying to shoot for the Mn Pro SoC price point. (likely the X1 die is smaller. Hence cheaper to make. Hence probably shooting for a lower price point. )
If Qualcomm deliberately sets out to make an "Orange" and Apple deliberately sets out to make and "Apple" then never going to get to an exact "Apples to Apples" comparison. There is no good reason why Qualcomm should ignore what AMD/Intel are doing in Windows and only focus on trying to fit the specs of an Apple SoC product.
The competition is in the Windows space. Apple just unilaterally nuked the x86 out of the Mac ecosystem. Nor are they going to consider X Elite/Plus as an option.
Closest? The M3 Max is a > 400mm^2 die. X Elite is not likely anywhere near the same zipcode as that kind of size. The X Elite is not trying to hit that price point for required memory package cost or SoC cost at all.
All that 'other, non-CPU-core stuff' coupled to the die matters. It influences just what the CPU design constraints are going to be ( allocated memory bandwidth, die are allocation, etc. )
Perhaps eventually Qualcomm may want to pursue a broader competition battlefront than with the X series, but for a first iteration the Mn Max isn't even on the map. The Mn Pro really isn't either this first generation.
Most of Qualcomm's comments so far have been more so geared toward moving these arch license cores 'down' the line up to even smaller dies ; not chasing 'big dies'. At least, it is likely to an alternative version that consumes less space ( compressed and even less single thread drag racing requirements placed on it. ) . Or a more substantively scaled down footprint E core. The SoC package size in a phone is relatively small. ( that is one reason why Windows on Arm has struggled to compete on performance when simply just reusing the same size dies as smartphones. )
Qualcomm isn't going to sell the laptops. That's where the slippery slope is. How much profit does Qualcomm skim out to cover their development (and acquisition ) costs and how much is left over for the system vendors (and Microsoft) ? Processor SoC costs are not 50% (or more) of laptop costs. Where RAM , Screen , SSD , etc costs also land will play a major contributing role too.
I suspect the die size of the X series is between the M3 and M3 Pro. Which means the costs will likely land between those two also. X series is on a more affordable node , but they don't have the same size amortization base to lay the fixed (and overhead) costs over. ( buying Nuvia was $1+ B right there. Don't have to make it all back in gen 1 but it was expensive. ) . No system vendor they have signed up is cancelling all of their AMD/Intel SoC orders so the volume is going to be capped.
The Windows laptops are going to be the more directly substitutable good here. AMD and Intel not launching a price war against Qualcomm is going to be more critical. Price relative to the x86-64 models is going to make or break this. If Qualcomm is selling notion that Windows vendors can slap on the "Apple Tax" on their models just because there is approximately a M3 equivalent in them ... that likely will run into some issues. The upside that Qualcomm should have is that there will be a range of prices; not a strict adherence to Apple's pricing models.
( Intel has some minimum requirements for things like an official "Intel UltraBook" or "Intel brand XYZ" stickers. Unclear how much Qualcomm is doing that. Or simply just trying to control that indirectly through SoC+"bundled chipset" pricing. )
And the MBA product price range is pretty board; roughly $1000-2200 . Once Apple's RAM and SSD capacity $/GB factors 'kick in', that opens the 'window' a lot. Similar forces are going to push the M3 Pro system out of the price overlap zone with the X Elite. And the M3 Max way out of the zone.
Hmm, ideally X1 Elite laptops should undercut x86 laptops, because there will be trust issues. I understand Qualcomm trying to position them as premium but I don’t think that will fly.
maybe this is what they were talking about? these DELL Inspiron scores are really, really bad...
a possible explanation would be that the DELL ran on battery and the SAMSUNG Galaxy Book4 Edge (Geekbench scores approx. 2700 / 13'000) was running while plugged into the charger..?
Last edited:
a more plausible explanation is that those Dell are either borked somehow or a faked entry. If go to the details of the scores. The first Dell is getting a single core Clang score of 2146 and a multicore Clang score of 4761. That is 12x more cores getting only a 2.2x increase in speed. That is really dubious. "Battery power" isn't going to explain that. ( Small chance something may be trying to hold the 'multicore' to just two cores (i.e., some Turbo only is good mania) . In that case, it would make sense to get about a 'two core" score. )
over on the Samsung Clang is 3335 jumps to 25112 which is a 7.5 That too is a bit quirky, but lots less dubious. The cores swamping the memory subsystem wouldn't be too surprising. Plus the "two core boost" is going to oversell the single thread score ( turbo will drop off once get all the cores going. ). Drop 15% from 3335 gets 2385 'non-Turbo core' score. That in turn leads to a 10x ratio which looks pretty reasonable.
[ and if apply the non-turbo reduction to the Dell it still looks silly; about 3x. Prehaps Windows has latched onto the 4 core cluster as being the limit of "multiple cores". If the Windows scheduler it trying to apply some legacy Big-little core strategy to the SoC just because it is "Qualcomm" then it is largely broke. Wrong, dated OS can likely lead to bad results. )
This qualcomm entry for a while back lists having "Windows Insider " as an Operation System.
Qualcomm ZH-WXX - Geekbench
Benchmark results for a Qualcomm ZH-WXX with a Snapdragon X Elite - X1E80100 - Qualcomm Oryon processor.
browser.geekbench.com
( Similar as Samsung 22009/3091 = about 7x )
given Microsoft has not formally released the version for these SoCs these all should be "insider builds". ( maybe , just maybe there is a release build in late April . However, folks throwing the "late 23 " , formally released Win 11 at these systems probably has tons of noise in it.)
This story has a X Elite reportedly 'disguised' as a 8cx Gen 3
Surface Laptop 6 with Snapdragon X Elite SoC confirmed via benchmark leak — rivals MacBook Pro with M3 and M3 Pro
The Surface Laptop is getting a powerful Qualcomm Snapdragon X Elite SoC this summer.
www.windowscentral.com
( also a Windows Insider OS version )
Last edited:
Qualcomm isn't going to sell the laptops. Lenovo , HP , Acer, Dell , etc will setll the laptops. Would likely be more so 'trust' in the system vendor since that who user will get support from.
That said, Dell has a decent chance of perhaps lining up with this. They tend to be hardcore Intel CPU company. Whatever Intel says Dell does to grab the biggest discount/access/support/etc . Mostly only Intel Chromebooks. Late to the Threadripper 'party'. etc. If there was a vendor that was committed to "save the limelight for Ultra 200 (Lunar Lake)" , then Dell would be a very likely suspect.
Some vendors may emphasize the 'lowball' path and others likely won't. I suspect there will be a broader set of offerings than when Qualcomm was charging a premium mainly driven by having a cellular modem built in. But they are not shooting for minimal bill of materials costs here. ( not slowest LPDDR5 they could have chosen, their own power management chips , not the 'smallest die size' could have chosen. etc. ) A substantive number of laptop users who use their device on battery most of the time will pay a premium for more battery life.
However, trying to throw 'low prices" at folks who are 'very hesitant' about moving off of x86 is a dubious path. Resistance to change is different than being constrained on budget.
Qualcomm may be hoping to attract customers because of the new features its superior NPU could bring. For example, rumors say that the new AI Explorer in the next Windows update would be exclusive to Qualcomm SoCs.
What is Windows Recall? Everything you need to know about Windows 11's new AI feature
Windows 11 is being supercharged with AI designed to enhance productivity and search across the OS.
www.windowscentral.com
Asus has just sent out a press release to the tech press about its upcoming event.
ASUS Announces Next Level. AI Incredible. Launch Event for its first New-Era ASUS AI PC
ASUS is set to unveil its groundbreaking new-era AI PC on May 20, 2024, in a must-see virtual event. This premium ultraportable laptop will offer innovative AI features and exceptional performance. Join us alongside speakers from Qualcomm and Microsoft, and get a preview of the ASUS AI PC lineup...
I wouldn't be surprised if other notebook manufacturers focus on AI in their marketing material too.
Last edited:
If the AI runs in the cloud, CPU features to enable it would make very little sense, I’d be surprised if they would go that way.
You are missing the boat here in a substantial fashion. Local " AI processor power" is going to be a requirement to get a certain level of "AI for Windows" certification.
" ...
Lewellen explained that Microsoft is focused on the customer experience with the new platforms. As such, Microsoft insists that Copilot runs on the NPU instead of the GPU to minimize the impact on battery life.
..."
Intel confirms Microsoft's Copilot AI will soon run locally on PCs, next-gen AI PCs require 40 TOPS of NPU performance
Let the TOPS wars begin.
" ... Bringing things back to AI PCs and their NPUs, Qualcomm’s Snapdragon X Elite chip includes an NPU running at 45 TOPS. That’s quite impressive considering the first run of Intel Core Ultra chips offer up to 34 TOPS with the CPU, GPU, and NPU pooled together, with about 10 TOPS for the NPU alone. AMD’s Ryzen 9 8945HS chip touts a total package limit of 39 TOPS, with 16 TOPS specifically from the NPU. ...."
What is an AI PC? Answers from Microsoft, Intel, and more explained
It's a term that you'll be hearing a lot as more manufacturers jump on board.
www.windowscentral.com
Intel (Ultra 200 mobile "Lunar Lake" ) and AMD ( Zen 5 ) should catch up to the 'TOPS lead' X Elite has by end of 2024. ( some of it is a bit hand wavy as X Elite is INT4 based . But Microsoft has laid down some goal thresholds to all of the Windows SoC vendors. Just like Pluton/Boot Security. )
Microsoft is going that way. In terms of making money on AI inference at very large scale, it makes way , WAY more sense to leverage the customers' equipment and electricity than to try to pay all of that at datacenter prices. ( Amazon Alexa was almost all cloud based and expensive datacenter costs were a contributing factor as to why it never made any money. ).
Additionally, talking about 'trust', how is sending all of your conversations with your assistant to a remote cloud building privacy trust? Will some folks be able to buy "CoPilot in the Cloud"? Probably. Will that be the same cost as 'local' inference? Probably not.
AI inference is not inherently necessary an any substantive way over the long term at all. According to rumors circulating on this site the vast core of Apple's strategy is not 'cloud' dependent. Some stuff will, but a very useful subset doesn't "have to".
Misdirection doesn't really address folks fear/FUD. They need some certain minimal levels of performance to offset the x86->Arm conversion overhead for whatever subset of legacy apps they probably also want to cling to.
Qualcomm tried the "hours more power , but substantively slower" route already. It didn't work.
As for "those Wattages", are they worse than the Intel ones? The AMD ones? Even if they are 0.5W per core better than AMD/Intel at 10 cores that is 5W. That will lead to an overall system battery gap. At 1W differential it is 10W.
Windows system vendors are already packing decent sized batteries into the laptop chassis.
ASUS Zenbook 14 (UM3406) review: This ultraportable AI laptop wants to travel with you, and it's got insane battery life to do it
Touchscreen OLED, AMD Ryzen 7 8840HS, and AMD Radeon 780M in a super lightweight AI PC.
www.windowscentral.com
All Qualcomm has to do is enable new boards to drop into the same chassis the systems vendors already have that using incrementally less power to get a battery life 'win'. It doesn't 'have to' drop into a Mac chassis and compete directly against any M-series system.
Overall system power consumption is substantively more than just the SoC. The screens contribute. The LPDDRn contribute. Going 'cheap' on screens and LPDDRn doesn't necessarily get you better battery life either ( OLED-dark mode + LPDDR5x probably better than IPS+DDR4 ) .
If X Elite/Plus can power isolate the 8 core cluster , then when users are doing low/minial work it could drop down to 2-4 cores at the base frequency is power saving was a focus. It is doubtful the wattages a couple pages back reflect anything that 'saving' mode would present. Time will tell how well the power management subsystem on these devices are, but decent chance the X Elite/Plus systems will gap the above by some hour(s) on 'same size battery' contexts.
P.S. Qualcomm isn't going to get an easy free ride here. AMD and Intel are working to close the gap for their mobile SoCs in 2024. But first round, there is a power consumption umbrella they can ship under. They would have been in substantially better shape if they had shipped end of 2023 and then had a 2nd generation ready by early 2025.
According to Counterpoint, 3 out of 4 laptops sold in 2027 will be AI-enabled laptops with advanced GenAI capabilities.
www.counterpointresearch.com
3 Out of 4 Laptop PCs Sold in 2027 will be AI Laptop PCs With Advanced GenAI Capability
We estimate that AI laptop PCs with the ability to run advanced GenAI applications will account for every three out of four laptop PCs sold.
There's a few reasons those could be so slow. Could be they were running other intensive stuff while running GeekBench, could be an engineering sample, etc. Here's an uncharacteristically poor M3 for example: https://browser.geekbench.com/v6/cpu/5894897
Doubt running on battery cuts performance so much, especially multi-core. It's only 800 points higher on multicore than my SQ1 on battery despite 4 more cores and a much faster chip by all accounts.
It would probably be half decent with linux. I would not have windows on one.
That’s like saying x% of machines will be “AV1” PC’s because the CPU has encoders built in.According to Counterpoint, 3 out of 4 laptops sold in 2027 will be AI-enabled laptops with advanced GenAI capabilities.
3 Out of 4 Laptop PCs Sold in 2027 will be AI Laptop PCs With Advanced GenAI Capability
We estimate that AI laptop PCs with the ability to run advanced GenAI applications will account for every three out of four laptop PCs sold.
It’s a fun marketing trick, just defining machines by hardware that is going to be included no matter what. That way you get a few more years of the hype cycle in stats.
It doesn’t matter if the machine was purchased to read emails, by definition it’s now “AI-driven” growth.
According to Counterpoint, 3 out of 4 laptops sold in 2027 will be AI-enabled laptops with advanced GenAI capabilities.
3 Out of 4 Laptop PCs Sold in 2027 will be AI Laptop PCs With Advanced GenAI Capability
We estimate that AI laptop PCs with the ability to run advanced GenAI applications will account for every three out of four laptop PCs sold.
That article is extremely dumb. “AI-enabled laptop” is a marketing term, it doesn’t mean anything. Slapping a mediocre matmul unit on a SoC does not make it “AI”. By that logic iPhones were “AI-enabled” for years. Also, Apple “being a dark horse” - they literally have an optimized StableDiffusion implementation for AS Macs, why won’t these “journalists” take a look? Apples ML performance is a known quantity. Unlike that of “AI-enabled” laptops.
The industry is slowly defining these terms. For example, according to Microsoft, an AI PC is a PC that can run Copilot locally and, to do so, is believed to need at least 40 TOPS.
Intel confirms Microsoft's Copilot AI will soon run locally on PCs, next-gen AI PCs require 40 TOPS of NPU performance
Let the TOPS wars begin.
According to CounterPoint's classification, current Macs could be considered AI basic laptops.
40 TOPS using what precision?
How are these categories defined? Base M1 should deliver close to 20-25 TFLOPs using FP16 precision.