Nothingness
Diamond Member
Do you know what laptop that was?
-
We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.
Discussion Qualcomm Snapdragon Thread
Page 81 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
poke01
Diamond Member
It’s from Lenovo. Don’t know the model
It's fine - great actually - for a part that's on N4/N4P and given where AMD/Intel are. Absolute power consumption topping out at 14W for a 4.2GHz part including full platform vs AMD and Intel where they have a 20-30% lead on them in terms of performance - that gives them both a notable performance gain (or responsiveness) and around a 20-30% energy efficiency gain still at those same power consumptions. This is also their peak.
And sure M3 would be 8-10W, and more for the M3 Pro/Max likely. But it's on N3B and Apple is now just iterating year in year out off a base, so. This is QC's first part with custom cores and was saddled with internal drag and lawsuits. Qualcomm just has to consistently be much closer to Apple than the other two, this really isn't complicated.
This is again just about the fundamental curves. They will likely improve the entire thing, but still keep something hitting 10-15W at peak, albeit with 25-40% more performance at that point. IOW, they will, but they're not going to take the power down to like 8W tops for a laptop part considering this is full platform. It doesn't mean the energy efficiency and slope of the curve won't improve.
poke01
Diamond Member
It is indeed possible this is the case, I admit it's way more likely X Elite is on N4 than 8 Gen 3. Ming Kuo originally claimed it was N4 as did others.
Yeah I don't know the re-spins would be able to literally take it from N4 to N4P. But a porting after an initial N4 run or something I could see given the design rules are constant.
RE: The Galaxy Book 4 Edge from the other day:
That thing has a 3K OLED display running at (dynamically refreshed I believe from 60-120Hz, but no advantage over 60Hz stuff), at 120Hz, and 400+nits of brightness, with a 55wh battery.
So 13H of normalized web browsing/streaming + other workload battery at full brightness with speakers on and responsiveness being quite good even at 2.5GHz is.... a really good result for the X Elite. If the battery were 75wh as is increasingly common for 13-14 inch laptops with OLED or higher resolution displays, it'd be more like 17-18 hours.
Granted I think this value will go down at 4GHz, but 3.4GHz should be able to maintain the same energy efficiency, just more performance.
That thing has a 3K OLED display running at (dynamically refreshed I believe from 60-120Hz, but no advantage over 60Hz stuff), at 120Hz, and 400+nits of brightness, with a 55wh battery.
So 13H of normalized web browsing/streaming + other workload battery at full brightness with speakers on and responsiveness being quite good even at 2.5GHz is.... a really good result for the X Elite. If the battery were 75wh as is increasingly common for 13-14 inch laptops with OLED or higher resolution displays, it'd be more like 17-18 hours.
Granted I think this value will go down at 4GHz, but 3.4GHz should be able to maintain the same energy efficiency, just more performance.
Nothingness
Diamond Member
So that's X1E-78 (which both Yoga Slim and T14s use), the version with no dual core boost. What the results show is that even with that limit, the X Elite power consumption is rather high (though that might also be preliminary bad power curves).
FWIW:
The TDP being 28W doesn't directly imply the ST is using 28W for example. They likely have different set points for it to give extra headroom beyond the boost. Even in Qualcomm's own A/B parts they did this, and it didn't match what their actual ST curves look like, where anything could fit under 20W. Which makes sense, especially if you wanted to have room for a dual-core boost (well, no boost on the base part but 3.4GHz I mean for each cluster doing ST) with the third cluster doing other stuff.
I also want to point out that the 106 score is a 3.4GHz score but the 65W mode hitting 133 for ST has to be above 3.4GHz. That's the 4GHz score. If this is the same laptop, then by definition it cannot be an X1E-78 part, because none of those will be hitting 133 in CB24 ST.
FlameTail
Diamond Member
I agree with you, but you are looking at from a technological PoV. What about the market? Intel and AMD are debuting their next gen laptop parts (LNL and Strix) very soon. Those will close the gap with Qualcomm. If the gap is too small then, consumers will feel no reason to buy a Snapdragon X device.
Then tell me why is the the 4.2 GHz/4.0GHz Boost (10W+) limited to only the top two X Elite SKUs? Every other SKU has no boost and tops out at 3.4 GHz (~7W).
And to top that off, those top two X Elite SKUs are also extremely rare. The vast majority of Snapdragon "X Elite" laptops have the X1E-78-100.
FlameTail
Diamond Member
Snapdragon X Elite and X Plus (3 8 TFLOP) GPU versions have appeared in Geekbench 6 GPU OpenCL list.
~20.000 points.
browser.geekbench.com
For context:
Apple M3 : 30000 points
Apple M2 : 26000 points
Radeon 780M : 26500 points
Radeon 680M : 25000 points
George 1060 : 30000 points.
~20.000 points.
OpenCL Benchmarks - Geekbench
browser.geekbench.com
For context:
Apple M3 : 30000 points
Apple M2 : 26000 points
Radeon 780M : 26500 points
Radeon 680M : 25000 points
George 1060 : 30000 points.
I am thinking about it from a market POV! You are crazy if you think Strix will get AMD's curve to match Qualcomm's. If it did, they'd be talking all about battery life, because that floor being 1/2 to 1/3 the power of AMD/Intel current gen and the rest of the curve giving a 20-40% performance gain is a huge energy efficiency advantage in addition to the lower idle power. So if AMD even cleaned up energy in ST by a substantial amount without cleaning up idle power (which, I think these are semi-related at a platform level anyways but) they'd still be running to the hills about that in the demo. And they have when they had advantages or improvements before, before someone claims "oh AMD doesn't talk about perf/W or GB6 composite ST" (LOL that's BS.)
Strix will have N4P (+5 or -10% power/perf respectively, at least for the chip alone, and these platform figures have much more in them that they have to fix) and probably some mild tweaks on it's side, and moving to 12C will most likely offer them some significant MT perf/W gains, but again, MT.
You have to actually think through this stuff instead of just assuming competitors will magically catch up. We saw the errors of magically expecting humongous gains just because of rumors, this lies in a similar plane.
What do AMD parts look like now? What do we know about AMD and what they've said about Strix at the conference? Not much about battery, not much about performance/W or energy either. The Asus demos claiming more performance at a given TDP are almost certainly both A) true and B) MT anyways which is unsurprising, moving to 12C will allow them to run at lower voltages with more cores for a given TDP (at least if it's high enough to where performance won't fall off, at like sub-10W they might have to disable a cluster).
But notice there's very little battery discussion and they sure as hell aren't going to show an ST perf/W graph or talk much about that - in fact the only ST info we really have concretely is a +5% ST in GB6 over the 4GHz X Elite.
Intel you are right though.
Intel at least showed a few things: LNC got 10-18% more performance iso-power with N3B vs Redwood on Intel 4, averaging about 12-15% I'd say, helped by both IPC and physical design changes and the node. Conversely looks like about 20-30% lower power at the same performance. That's core alone. What they said for the part itself vs previous gen is "similar ST at half the power" which, I suspect is referencing the peak ST of MTL or something from LNC. So the package/SoC is making a difference too beyond that 20-30%. It could also be referencing the LPE Cores being able to roughly match a P core at twice the power (taking the full SoC into account). Notice how they're talking about like SoC power or package power often.
Now again, it also depends on how they measured here and the power delivery/package part going down could take them even lower, but even then, if Intel is getting the same ST as Meteor Lake at 50% less platform power, it would put them in line with or even slightly lower than Qualcomm. I expect it to be pretty similar and Intel to switch out E Cores and P cores with the P cores doing worse below 5-6W (so you'd want to measure both).
But one thing you're missing about Lunar Lake, or two things really.
1) Intel is using N3B and a bunch of transistors, it's a 140mm^2 part + a 40mm^2 IO die on N6 (though we can excuse the N6 part I guess) and they won't even match an M3's ST perf/W curve, there is no way.
Qualcomm has a lower cost part, it's N4P (or N4, whatever) and a 170mm^2 die. We already see Lenovo Slim laptops for $1280 with 32GB of LPDDR5X and 1TB of storage with an OLED, or $1200 starting. Alternatively X Plus laptops with 16/512 and a nice 400 nits 2.5K LED display for $1099, which I seriously doubt you're going to find right off the bat with Lunar Lake, and by the time you do, the QC versions are going to be on sale or cheaper. If you want great battery life and responsiveness, QC will still have very competitive offerings even before Purwa.
2) Besides that, Qualcomm will *probably* give you more MT at pretty much any part of the 10-12C continuum. We don't know what LNL's MT is just yet, but the rumored + 50% in CB23 and GB5 at 17W (unclear if that really means PL1 throughout the test) doesn't really make me concerned, because A) QC/Apple stuff does better on CB24 after NEON support, and the performance they have over H units with more cores than the U units in CB24 is sizable enough.
With 1 and 2 in mind, Qualcomm has the only part that can compete with Lunar Lake and Strix Point both. If you want great battery life and ST + MT, Qualcomm has that covered and with a lower cost structure.
Ghostsonplanets
Senior member
Seems like we're about to see a performance and battery champion. That's basically a paradigm shift when interacting with a laptop.
Now one could drive a high-end laptop with amazing features active and don't worry about battery life. OEMs can deliver amazing laptops and new form-factors without being constrained by the CPU.
The future iterations of Snap X series will certainly be something to see.
FlameTail
Diamond Member
Some other minor upgrades that X2 Elite could/should bring:
• PCIe Gen 5 (X Elite has only PCIE Gen 4).
• Support for higher resolution/higher refresh rate external displays (X Elite can support 3 external displays but only at 4K60).
• Improved VPU (X Elite's VPU reportedly tops out at 4K120 decode, 4K60 encode).
For context, Lunar Lake;
• Has PCIe Gen 5
• Can drive three displays total at 8K60.
• Can do 8K60 encode/8K60 decode
*VPU : Video Processing Unit
• PCIe Gen 5 (X Elite has only PCIE Gen 4).
• Support for higher resolution/higher refresh rate external displays (X Elite can support 3 external displays but only at 4K60).
• Improved VPU (X Elite's VPU reportedly tops out at 4K120 decode, 4K60 encode).
For context, Lunar Lake;
• Has PCIe Gen 5
• Can drive three displays total at 8K60.
• Can do 8K60 encode/8K60 decode
*VPU : Video Processing Unit
Unclear what you mean here. I said:
"This is again just about the fundamental curves. They will likely improve the entire thing, but still keep something hitting 10-15W at peak, albeit with 25-40% more performance at that point. IOW, they will, but they're not going to take the power down to like 8W tops for a laptop part considering this is full platform. It doesn't mean the energy efficiency and slope of the curve won't improve."
They'll keep those targets because they can without just getting sloppy with it. The whole curve will still improve in terms of power/performance.
You don't seem to actually understand why those SKUs exist. They exist because of binning and because Qualcomm decided they could push clocks some at acceptable power tradeoffs - the X Elite 78 and X Plus SKUs are lower quality and couldn't be pushed due to either inability or unacceptable leakage and power tradeoffs at those frequencies.
That's why I'm saying they'll do it again next time, albeit a node gain and IPC gains along with probably other cache changes will "push" the performance upwards at any given power, and also push the floor SKU clocks upwards. This is why I was mentioning what Apple could do since they don't really frequency bin.
So next time the entire curve will improve, and the base SKUs might be 3.8-4GHz with N3P, and permitting binning and timing constraints for the core, the new top SKUs might be at 4.4-4.6GHz.
The reason AMD/Intel stuff is higher clocked and can get fine bins still is less because of like core size and timing at this point (Lion Cove is still plenty wide and Zen 5 added some heft) and more because of the transistor choices, which impact the frequency you can hit and also the yields for those frequencies. See why Qualcomm's bottom SKUs are around 3.4GHz and similar for Apple on N4/5, whereas AMD can easily assure 4.8-5.2GHz on N4/N4P mobile parts.
But of course that stuff also consumes way more power anyways. Qualcomm is not going to do that. Think of what they do as opportunistic more than like deliberately trying to get crazy clocks, which is why it'll take N3E/P for them to get 3.8+GHz as the standard floor, because of their library and transistor choices. One thing that annoys me is the clowning about clocks from Armlets and then the counter mockery over M3/4 pushing them per se etc after people claimed Apple wouldnt do that blah blah.
Clocks aren't bad, it's good to ride out node gains, just without deliberately changing core designs and inflating the f out of area and power to chase more peak frequency within certain yields.
Rare for now. 4GHz X Elite is in the XPS, Galaxy Book, Surface Laptop, so more likely just very SKU specific which is a good use of it.
But again this also goes back to what I was saying: these are bins. It could also be Qualcomm choosing to put more into the X1E-78 depending on demand too, but if you're saying "they're trying to keep power down", no, lol, if they could ship every SKU as the 4GHz one with that exact curve, they would, and they'd let OEM's lower the clocks if they so please. That power tradeoff from 3.4 to 4GHz is nothing like the BS AMD and Intel do where the slope is nearly flat from 10-25W and at a slow incline.
Yep. I couldn't believe it was 55wh, I was assuming it was 70-75 and that would still be a great result. It's honestly ridiculous.
Also, it was limited to 2.5GHz, and yet he said responsiveness was fantastic, which makes a lot of sense if we're being honest. That 1800 GB6 ST is still usable for sure, and if they ramp often or fast enough to it it makes perfect sense it still feels so responsive.
Now, so the battery life would decline at 4GHz - but not really 3.4GHz which is really still on a great tradeoff point. But even if power declines by a bit when using the 4GHz setting, it's not the same tradeoff AMD/Intel used to make from their own platform curves (though it is like almost twice the power for 30% more performance over the 3GHz range, so a decline for ST) and idle is still fantastic as far as we can see, so this is really overall just a great showing.
It reminds me of something I pointed out early on: Even if you limited the clocks on these chips to save energy, you're still in practice going to get so much more responsiveness and performance than doing the same to an Intel/AMD laptop currently and it's not like that's some uncommon thing either. Also, it's not like the Lunar Lake strategy is "let's give them full ST constantly" either (not saying it won't be a good part but just that it's funny people would clown and ignore somebody saying this thing is responsive at 2.5GHz and really efficient, when Intel's own strategy is taking things even further and with much less to show for on MT, using a better node too).
I'm honestly kind of floored by how good the early reports are. The fact that Oryon V2 is coming with apparently a big IPC upgrade, and a substantial node upgrade (should push clocks up for the whole range) and maybe other tweaks like real e cores is going to take this to the next level if all pans out.
FlameTail
Diamond Member
I understand why. The 4 GHz+ parts exist as flag carriers. They pushed the clocks to the limit of the silicon, as you can see in this power curve how the power shot up by 50% for 3.8 -> 4.2 GHz
Again it all hearken backs to the rumours, that X Elite was really supposed to launch in 2023 and compete with M2, but it got delayed for various reasons. Qualcomm had to pull several tricks, such as 4.3 GHz boost and running it in Linux, to match the M3 in GB6 ST.
They really need to reduce the ST clock difference between the topmost and bottom-most SKUs. 3.4 GHz -> 4.2 GHz is a substantial 23% difference. This kind of difference is never seen in Intel/AMD/Apple parts.
Speed demons (x86 cores) vs Wide Monsters (ARM cores).
LOL.
Assuming a 25% IPC gain and going through some N3P boosts to clocks across the SKUs (and also knowing the core structure itself already allows them to hit up to 4.4GHz at at technical level).
Doing the math, a 4GHz (about 17%) clock gain over 3.4GHz for the base X2 parts would have them at 3566 GB6 ST.
For the existing 4GHz part going to 4.4GHz, you'd be at like 3850 GB6 ST.
And for 4.2GHz, if they went to 4.6GHz for almost a 10% gain and they could get about a 4038 GB6 ST (again adjusted from their existing score at that frequency).
The 5GHz rumor I doubt they've made enough changes to allow them to hit that frequency without blowing power too far, but I can see a 4.4-4.6GHz top SKU setup on N3P. Maybe it is true though, and in that case: 4389 GB6 ST.
Regardless though it'd be an absurdly competitive part assuming the top power didn't change much and this is just pushing curve up which seems likely.
Too much could happen again, I don't have hopes up, but given GWIII's comment and the circumstances it's believable their next part would be this good.
Yes ofc. But also that tradeoff, while bad, won't actually be made for most parts and is still again far better than the tradeoffs made in most cases. It's not worth going nuts over.
FlameTail
Diamond Member
Major IPC increase doesn't come free, without a power increase.
If you are using the entirety of the node gains to increase the frequency, then you have no more node gains left to cover for the power increase brought by widening the core.
This is why my opinion is that Qualcomm should should not jack up the clock speed for Oryon V2, but do a clock regression + major IPC improvement. This way, they can deliver a major performance uplift without increasing the power envelope from 15W ST (or maybe even reduce it).
Yep 😉
That depends on what you do, though. And it also depends on what goes on with cache and physical design changes. You're right, but you're just missing that the process gain for the frequency iso-architecture isn't the only change that will occur in tandem to + IPC and + some extra power. If the added IPC took power up 10-15% for the core throughout the whole frequency curve before node changes, but extra L2 and SLC took data movement down, they could be just fine. I know what you're thinking here and I don't disagree but it just depends.
You are assuming the node changes are the only form power reductions can come in. Even with Nuvia I bet they have some phydes tweaks and core cache changes. For ex: Apple with the A16 took frequency up by another 10% despite moving from N5P to N4, but with 33% more L2 and some other tweaks they brought core power (iso-performance) down by 20% per their own presentation, and in practice after that and admittedly LPDDR5 (which tbf I think helped for the platform power) they got GB5 scores up at around the same power.
FlameTail
Diamond Member
Lol, are we so bored that are already talking about X2 Elite, before X Elite itself hasn't even properly released😛
Reviews are 3 days away. Till then...
I think we can go back to that Oryon architecture disclosure and discuss about some of the finer details:
www.anandtech.com
Reviews are 3 days away. Till then...
I think we can go back to that Oryon architecture disclosure and discuss about some of the finer details:
Qualcomm Snapdragon X Elite and X Plus: Specs, release date, benchmarks, and more
The next big push for Windows on Arm is coming, with Oryon cores.
FlameTail
Diamond Member
For instance, the Adreno X1 GPU architecture does support Ray tracing but only Level 2.
blog.imaginationtech.com
Here are the six levels of Ray tracing, as defined by Imagination:
Introducing the Ray Tracing Levels System – and what it will mean for gaming - Imagination
Not all ray tracing solutions are created equal. Learn about the Ray Tracing Levels System here.
blog.imaginationtech.com
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 25K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 24K
-
-
