Even rumored Switch 2 Soc has combination of Ampere Gpu and Cortex A78 cores, while not ideal for high perfomance laptop, its good combination for Tablets and some ultrabook which focus on batterylife.MediaTek did demo their ARM SoC and RTX 3060 in Chromebook, so everything is possible...
Microsoft: “It’s exciting to see NUVIA join the Qualcomm team. Our partnership with Qualcomm has always been about providing great experiences on our products. Moving forward, we have an incredible opportunity to empower our customers across the Windows ecosystem,” said Panos Panay, Chief Product Officer, Microsoft.
If it’s true that MS went with NVIDIA: hasn’t Microsoft seen samples of the NUVIA laptop SoC, Oryon? Qualcomm has long hyped how revolutionary their next SoC is vs incumbents at Arm.
Does Microsoft believe Cortex-X4 / X5 are close enough to NUVIA’s Phoenix core? 🤔
In the NUVIA acquisition, Panos Panay was the first in Qualcomm’s PR:
The fact that Microsoft apparently got tired of waiting on Qualcomm to deliver something I think is telling after they apparently were this patient this long (but perhaps that's why, disappointment with what Qualcomm has offered thus far and further delays), and doubly so that they apparently went with NVidia who also has had a problem delivering on ARM. Either NVidia had an ARM design spec'ed out already that reasonably met what Microsoft is looking for, or there's something else going on (NVidia touting leveraging them for AI). Or maybe Microsoft being Microsoft and giving up being burned by one to get burned by NVidia yet again.
It makes one wonder if Qualcomm was intentionally trying to sabotage the market, they're incompetent, or maybe something else. Which with Qualcomm ditching Intel, could Intel be the reason for this, either intentionally or not (meaning, either their fabs are just struggling that much with new processes, leading to another delay for Qualcomm or they intentionally were jerking Qualcomm around to stymie ARM on Windows).
Guess if we see who this other party is, it might offer some clue. My guess is it'll be Samsung, hopefully with AMD GPU (which would be a much better fit for that type of chip). Also, hopefully it might mean that they're going with fairly stock ARM CPU cores, because NVidia has a bad track record with custom ARM designs (unless in their buildup to the ARM acquisition they were already working on something). Common ARM CPU and NVidia and AMD GPUs would likely massively help accelerate ARM on Windows development. The GPU commonality with existing Windows should help provide mature graphics drivers (and bolster some markets that appear to be developing like Steamdeck like gaming focused portables). Or maybe some Microsoft spec ARM design which they then have for Nvidia and possibly others to produce.
Random thought, could hybrid design go x86 for performance and ARM for efficiency cores (although I think there it would be more about density, cramming a lot of cores in the smallest area).
CPU | Geekbench 6 / 1T points |
Intel i7-1265U (Golden Cove) | 1886 |
Qualcomm SD 8Gen2 (Arm Cortex-X3) | 1877 |
AMD 6900HX (Zen3) | 1966 |
I think at this point MS has seen the writing on the wall with Apple's ecosystem who are offering the following:
1) Advantage in CPU performance vs Qualcomm's disastrous efforts, there is not much to say when the gap is almost 2x in GB6 ST, not really same class of devices.
2) Incredulous advantage in GPU performance
3) Very competent x64 emulation, helped by CPU customization
4) Apples efforts at making use of Steam Deck game emulation driven advances in GPTK. I think if they were serious about it, they could be running a lot of games very capably. Things will only get better there, as they are using industry momentum in graphic framework translation on Linux.
5) very competent SoC overall with acceleration of a lot of things relevant for today like video, image, voice, AI etc.
Clearly MS felt that they need SoC with seriuos graphic's/video codec/acceleration IP and I think the main players are:
1) Samsung with that AMD GPU IP collab they've announced that felt out of place in phones
2) X party with Nvidia's IP
Who is X? Probably Mediatek.
They might've fumbled something if those new Nuvia leaks are accurate but it's a world away from 8Cx Gen 3.
The Nuvia leak has GB6 at 1700 and when I look at the GB site 8cx Gen 3 is 1500. That's an improvement but hardly "a world away".
The Nuvia leak has GB6 at 1700 and when I look at the GB site 8cx Gen 3 is 1500. That's an improvement but hardly "a world away".
Just fot reference current SD8 Gen 2 for smartphones scores ~2000 Sigle core in Geekbench 6 ( on Android with no emulation obviously ). And in a few months we will have the next one with at least another 10% improvement in SC.
The only way to catch up with Apple's ST performance is to devote the die size to it. The Avalanche cores are just larger in every way. They have 3x the L1 instruction cache, 2x the L1 data cache, and 4x the L2 cache. Although not usually relevant for ST performance, the M2 also has nearly twice the bandwidth to main memory. It's also a wider design with a ridiculously large reorder buffer.
The Cortex-X4 is going to be interesting. It has beefed up the execution and allows for double the L2 cache. ARM's performance slides aside, I'm looking forward to seeing the X4 in different L2 cache configurations. That should give us an idea of how cache-starved the core is (which is where it really lacks compared to Apple's performance cores).
From SemiAccurate, a series. Yes, it's paywalled and this is effectively a blurred line between rumors and news but Charlie's credibility isn't exactly MLID-tier and we discuss those things, so this I think is worth a thread. There are four posts in total, intertwined effectively:
I) MS made a deal with QC on WOA notebooks
II) QC lost that socket - or two. (Presumably Surface-related)
III) In Qualcomm's place, someone unexpected won that socket/contract - who?
The curious finale in IV, independent of the firm that won that socket Qualcomm lost (may not be an Arm-based replacement though) there's a new entry to the WOA space.
IV) "A new player enters the ARM laptop SoC space"
-----------------------------------------------------------------------------------------------------
I) and II) are much more difficult to guess about and I think generally less interesting other than that Nuvia delays are possibly if not probably in play and QC's sloppiness hurt them, maybe even a pricing issue.
On III):
Assuming it isn't restricted to WOA here, to me this isn't Intel but probably MediaTek or plausibly AMD, because of "given the company's ambitions".
That is very straightforward here with regard to MediaTek's interest in something competitive for WOA as they've voiced publicly two if not three different times (maybe more) since 2020. Alternatively, for AMD everyone knows they're interested in growing their mobile pie - which is wise, because DIY crap is frankly kind of irrelevant.
On IV):
Putting aside Charlie's stylistic bitterness, this has to be Nvidia or Samsung. The mention of Apple alongside the seismic shift makes me think so.
For Samsung's client/consumer division, they'd have to use stock IP obviously but they absolutely could ask for and buy beefier Exynos chips to throw into laptops without much issue (besides drivers) with their industrial scale and since they are an endpoint laptop vendor. MediaTek has to sell to someone directly, and has a tougher time with the chicken and egg game. I don't know it's likely, but it's possible, I guess.
However, I think it's far more likely to be Nvidia. Stock cores + GeForce IP on a recent process node could be popular and they'd have a reciprocal effect on developer interest.
Those margins will decline somewhat long term, it’s inevitable. Anyways, this is a curious line of reasoning and I’ve explained why I don’t expect it to govern their behavior here because firms aren’t unitary actors. They sell fricking Orin SoC’s today along with dGPUs for laptops and there are other payoffs for Nvidia in expanding their revenue streams — namely in amortization of IP development. The article is indeed about Nvidia from what I gather, they paid for the Surface slot after 2024.There is a famous quote from supermodel Linda Evangelista, saying she does not get out of bed for less than $10,000.
With NVidia pocketing profits well over $10,000 per H100, it would be curious if NVidia was interested in market where it is an uphill battle to make $10 per chip.
The Nuvia leak has GB6 at 1700 and when I look at the GB site 8cx Gen 3 is 1500. That's an improvement but hardly "a world away".
RE: Doug’s claim:That’s not 8cx Gen 3. You don’t know what you’re talking about.
It was GB5 at 1700-1800 for starters and 8Cx Gen 3 is at 1100. That’s a huge difference, and it’d be on TSMC N4, not Samsung 5NM.
The MT is massively improved as well.
I’d also like to add the L1/L2/L3 structure on the 8Cx Gen 4 suggests the actual energy consumption will punch above it’s weight if you solely looked at IPC or the node. Easy way to reduce energy consumption by keeping data closer to the core.
Yeah the actual improvement over the current Arm IP is now low or insignificant which is why it’s disappointing. But it is still a major upgrade over the existing WoA SoC performance which is a separate issue.Just fot reference current SD8 Gen 2 for smartphones scores ~2000 Sigle core in Geekbench 6 ( on Android with no emulation obviously ). And in a few months we will have the next one with at least another 10% improvement in SC.
That’s not 8cx Gen 3. You don’t know what you’re talking about.
It was GB5 at 1700-1800 for starters and 8Cx Gen 3 is at 1100. That’s a huge difference, and it’d be on TSMC N4, not Samsung 5NM.
If current Snapdragon 8 gen 2 would wipe the floor with their upcoming Nuvia design, why would they even bother?Are you sure he doesn't know what he talks about?
8cx gen3 is 1100 for GB5 and 1500 for GB6.
Are you sure he doesn't know what he talks about?
8cx gen3 is 1100 for GB5 and 1500 for GB6.
The Nuvia leak has GB6 at 1700 and when I look at the GB site 8cx Gen 3 is 1500. That's an improvement but hardly "a world away".
I completely misread your post: I thought you were saying the 8cx gen 3 was 1100 for GB6 and 1700-1800 for GB5. Hence my two linksI am actually, I'm very confident.
—-
"Rumor says about the overall situation of Nuvia-based 8CX Gen4 development in Qualcomm.Initial goal (January): GB5 Single 1900pts; 2000pts at 4GHzMay: Lowered goal to 1800 pts in GB5 SingleResult: 1600 @ 3.4 GHz, 1775 @ 3.8 GHzSOURCE: https://gall.dcinside.com/galaxy/998869 (in Korean)"
—
Now, the core is disappointing seeing as it doesn't match Firestorm/Avalanche IPC nor the Cortex X4 - so it's entirely plausible as we're seeing that things may fall short on performance but that's within ballpark of both Nuvia estimations or just sanity.
—
Will it be enough to compete with Strix Point or Intel's new products, or the M3 (assuming market crossover)? Don't know yet, also depends on how quickly they get a second model off and also the energy efficiency/battery life of these products, IMO. But lol, it's not THAT bad RE: 14% better than 8Cx Gen 3.
CPU | "IPC" (Pts / GHz) | GB5 1T Pts | Frequency |
Cortex-X3 (SD8G2) | 438 | 1,398 | 3.19 GHz |
Cortex-X2 (SD8G1) | 409 | 1,309 | 3.20 GHz |
Cortex-X1 (8CXG3) | 362 | 1,087 | 3.0 GHz |
Phoenix A | 467 | 1,775 | 3.8 GHz |
Phoenix B | 470 | 1,600 | 3.4 GHz |