Nuvia: don’t hold your breath

Correction:

It must be noted that these numbers are shipments, not sales!
Canalys told TechRadar Pro, “As this was the first full quarter of shipments for Snapdragon X Series PCs, we saw sequential growth of around 180% compared to Q2 2024.
It must be noted these numbers are shipments, not sales
Click to expand...
Using the 180% QoQ figure and 720,000 shipments figure for Q3, we can work out that the shipments for Q2 was 400,000.

ShipmentsQoQ
Q2400,000
Q3720,00p180%

This means Q2 + Q3, they have shipped 1.1 million Snapdragon laptops!

That looks even better.

I think they are on track to ship 1 million in Q4, and finish 2024 with 2 million+ shipments!

That's exactly what Ming Chi Kuo predicted:
The X Elite and X Plus chips, used for Windows on ARM (WOA), will reach about 2 million unit shipments in 2024
Click to expand...
medium.com

Two new growth drivers for Qualcomm / Qualcomm的兩個新成長動能

The SM8750 (Snapdragon 8 Gen 4), which will enter mass production in 2H24, is expected to be priced 25–30% higher than the current flagship…
medium.com medium.com
 
The Flame said:
Correction:

It must be noted that these numbers are shipments, not sales!

Using the 180% QoQ figure and 720,000 shipments figure for Q3, we can work out that the shipments for Q2 was 400,000.

ShipmentsQoQ
Q2400,000
Q3720,00p180%

This means Q2 + Q3, they have shipped 1.1 million Snapdragon laptops!

That looks even better.

I think they are on track to ship 1 million in Q4, and finish 2024 with 2 million+ shipments!

That's exactly what Ming Chi Kuo predicted:

medium.com

Two new growth drivers for Qualcomm / Qualcomm的兩個新成長動能

The SM8750 (Snapdragon 8 Gen 4), which will enter mass production in 2H24, is expected to be priced 25–30% higher than the current flagship…
medium.com medium.com
Click to expand...
Slight correction:

It’s 257,000 units shipped Q2 (100% growth is double, 200% growth is triple so 180% growth is 2.8x). I get that wrong all the time.

Now close to a million units shipped in that time frame is not bad, but that’s shipped not sold and unfortunately doesn’t change the headlines much. Still, on the positive side that can serve as the catalyst for better growth for Elite V2. But I maintain my position laid out in my previous post on the list of things Qualcomm (and MS) needs to do to make Windows on Arm a competitive platform and why the Snapdragon Elite wasn’t as successful a launch as it could’ve been.

However I should also state that while I think the Elite as an SOC would’ve fared better in June 2023 than June 2024, given the state of Windows on Arm in 2024, it’s possible that Windows on Arm in 2023 would’ve held it back even more. So an earlier release might not have been a panacea.
 
The Flame said:
X Elite does have the PCIe lanes to support a dGPU. That's what Qualcomm claimed in October 2023, IIRC.

According to rumours, and corroborated by die shots, X Elite has;

2 PCIe G3 lanes for WiFi/BT
2 PCIe G3 lanes for 5G Modem
4 PCIe G4 lanes for SSD
8 PCIe G4 lanes for dGPU
Click to expand...

Looks like it has exactly what it needs for the segment, and no real issues putting a high end GPU. It'd be absolutely hilarious to pair this thing with a 400+W monster GPU. :)

Although I can already hear the cries of the wider gaming nerd community that it "only" has 8 lanes for the GPU and "needs" 16...
 
Jimmyjames said:
More benchmark shenanigans. This time from OnePlus. It’s not their first time. Usual method.

View attachment 32880
View attachment 32881
View attachment 32882
Click to expand...

Yeah. Like I said this is OEMs, not QC. Still sucks and pretty egregious. They should keep ramping the same without app profiling and just ship the light performance mode by default. The 6-7.6W range isn’t worth the trade off at all. Frankly I’d keep it at about 3.3-4GHz which seems to be the best big core tradeoff range while still getting good performance.
 
Cmaier said:
someone want to summarize? I hate watching videos for this kind of stuff.
Click to expand...
Oryon is a good core, X Elite as a product in laptops is meh. Battery life is often bested by Lunar Lake and of course Apple Silicon. x86 32 bit app performance is diabolical. The second screenshot is a little misleading in terms of the captions. He continues “…about the actual performance of this PC product."
1732822830906.png
1732822839353.png
 
Jimmyjames said:
Oryon is a good core, X Elite as a product in laptops is meh. Battery life is often bested by Lunar Lake and of course Apple Silicon. x86 32 bit app performance is diabolical. The second screenshot is a little misleading in terms of the captions. He continues “…about the actual performance of this PC product."
Click to expand...
oh! wasn’t expecting that.
 
The big thing is actually just that it corrected the Snaodragon Spec Test issue: It now looks much more like Qualcomm’s actual graphs with the curves (tho different tests, similar power floor).

IMG_7043.png



This is the previous one. It’s BS, and Intel/AMD guys went nuts about it as if we were astroturfing pointing out Linux support was lacking (Geekerwan did not realize that for the X it seemed) and that plainly the floor was too ridiculous lol.

But good on him for correcting and measuring again.
 
X Elite has problems due to Windows on Arm & awful GPU.

But the CPU cost and engineering relative to anything from AMD and Intel is…. Really good which is pathetic given it’s a disappointment and they obviously made rapid gains onto Gen 2 and popped out an e core as well.
 
Just kind of shows how bad Intel and AMD are at this stuff; even Intel taking low power seriously spent much more than Qualcomm and couldn’t really bring a result that’s better substantially on battery life, Qualcomm’s immediate successor is far ahead, Intel’s margins are in trouble etc. And next time they’re going to be going back to more chiplets, optional PMIC use for OEMs, no MOP, etc.

A good thing I can say about AMD is they’re doing better by virtue of keeping this simple and killing it in the datacenter, also not in the margins pit Intel is for mobile so.
 
Ironically I bet AMD’s SoundWave (code name) Arm APU coming for Surface with Cortex cores and probably some investment into proper power delivery will be a much more credible effort than LNL now that I think about it lol
 
Jimmyjames said:
Oryon is a good core, X Elite as a product in laptops is meh. Battery life is often bested by Lunar Lake and of course Apple Silicon. x86 32 bit app performance is diabolical. The second screenshot is a little misleading in terms of the captions. He continues “…about the actual performance of this PC product."
View attachment 32904View attachment 32905
Click to expand...
Yeah his analysis was similar to mine but way more brutal and frankly exasperated at the end, especially by the compatibility issues. Apple really did themselves a favor by culling 32bit apps a long time ago on both iOS and macOS to motivate the switch before transitioning to everything to ARM. The 32bit translation performance must've been what Charlie Demarjian was referring to when he was talking about the performance being dog shit 2000s Intel Core-level under certain circumstances (and he was blaming MS not Qualcomm). Battery life under light loads is very good for Lunar Lake, most reviewers emphasized how good Lunar Lake was under those conditions (probably the E-cores and all the investments into low power delivery), and seemingly better than for Snapdragon Elite in that regime - though Qualcomm should be at a massive advantage under heavy, and native, loads. But overall, that's still the problem. Without a killer reason to switch architectures the Elite has to compete on price and even though supposedly the Qualcomm chip is cheaper than the competition to produce and for OEMs to buy, so far the prices on the end devices are still high - which I guess is on the OEMs?

I would disagree with his contention that Qualcomm should've focused on netbooks and putting mobile chips into them (though obviously the lack of 5G modems for the Elites is an odd omission given the Snapdragon pedigree and that it's Qualcomm), but I do agree that Qualcomm's first fully custom SOC foray was late and oddly structured in design. At least ... it's not how I would've approached it. But then again, they didn't have a small core at first so my preferred approach of going thin and light first with an M1-like SOC might not have worked either. Anyway, Qualcomm has to step up Elite V2 and at least now they have E-cores which should enable better SOC designs.
The Flame said:
Important screenshots:
View attachment 32914View attachment 32916View attachment 32917
View attachment 32934
View attachment 32918View attachment 32932View attachment 32920View attachment 32921View attachment 32922
Click to expand...

Jimmyjames said:
So there is the Geekerwan Lunar Lake review discussion on Reddit.
Click to expand...

So it's good that running SPEC under WSL made for better results than the original ones. Hopefully they are now closer to reality. The old SPEC results, especially SPECint made little sense. Elite V1 still doesn't perform was well as I would've thought but it's a heck of a lot closer and certainly would look all the better if Lunar Lake didn't exist. I was intrigued that Geekerwan made such big deal about the possible energy savings of memory on package for the Apple chips and Lunar Lake versus the memory on motherboard for the Snapdragon Elite, especially at the low end of power. Of course Intel is said to be abandoning that in future chips.

I'd say that WINE/GPTK having as good if not better translation performance for games than MS' prism is surprising, but it really isn't. It should be. But it isn't. Then again, the WINE project has been around a very long time and is very mature (and while Apple can't take credit for that, the GPTK project as been a huge boost as well which Apple can take credit for). But even so, MS has been trying to make WoA a thing now for over decade and they simply haven't got their shit together. Translating these titles for Apple products means having to do extra work and still the software/hardware is enabling as good as if not a better experience than staying on Windows albeit with a different architecture.


Artemis said:
The big thing is actually just that it corrected the Snaodragon Spec Test issue: It now looks much more like Qualcomm’s actual graphs with the curves (tho different tests, similar power floor).

View attachment 32940


This is the previous one. It’s BS, and Intel/AMD guys went nuts about it as if we were astroturfing pointing out Linux support was lacking (Geekerwan did not realize that for the X it seemed) and that plainly the floor was too ridiculous lol.

But good on him for correcting and measuring again.
Click to expand...

Artemis said:
X Elite has problems due to Windows on Arm & awful GPU.

But the CPU cost and engineering relative to anything from AMD and Intel is…. Really good which is pathetic given it’s a disappointment and they obviously made rapid gains onto Gen 2 and popped out an e core as well.
Click to expand...
Ha! You and I are typing at the same things at the same time and you're getting ahead of me by just a couple of minutes. :)
 
Last edited:
IMO the Linux issue wasn’t IPC so much as the power floor which riled me up because it jaded the curve more than anything else. I don’t think IPC even changed or aggregate performance really by much
 
Also agree X elite CPU good on paper esp considering cost, bodes bodes well for future, but yeah I mean the SoC as a whole… Arm isn’t there yet for a $1000 laptop to make sense for most with this
 
If he’s claiming a pure magical scalar load might be faster at 5.1GHz, problem I have with this within these variations is that’s not how this works with real workloads and with modern prefetching, cache, branch prediction etc.

Like, no one wants a 6GHz Cortex A720 vs a 3GHz Apple Firestorm lol
 
Back
Top