Question Qualcomm's first Nuvia based SoC - Hamoa

[poke01]

Qualcomm's working on a 2024 PC chip codename "Hamoa" with up to 12 (8P+4E).

Said to the same cache layout as M1. large private L1$, per-cluster L2$ (cluster = 4 cores, 12MB for every cluster) and a lot of LLC.

source:

https://twitter.com/x/status/1589405172979339264

 

[SpudLobby]

Intel knew that there was nothing in the world they could do to prevent Apple from dropping x86 Macs. Apple would have switched even if Intel was coming out with better CPUs than what Apple had, though Apple having something better or at least equivalent obviously smoothed the transition for them.

Intel knew that Apple's ability to steal x86 Windows market share was extremely limited, because they run a different OS, and they only compete on the high end not in the mass market. There would be little to gain by 'responding' to Apple because they cannot ever get Apple to make x86 Macs again and will be just as limited in their ability to steal Mac share as Apple is in its ability to steal Windows share. The equation changes if they believe Qualcomm will have a chance to make inroads into Windows market share (I'm still skeptical)

Intel’s Lunar Lake is nakedly an acknowledgment and response on the note of ultramobile from a company that has consistently failed to take mobile — even what they have left to defend — as seriously as it should beyond god awful LKF experiments and throwing wattmaxed blobs that never belonged in mainstream laptops.

Apple in the end obviously would use their own I agree.

Anyway, it’s pathetic Intel’s “we’re kinda taking this really cereal this time” LNL won’t be here until late 2024 early 2025. Why wasn’t that there in 2020 in principle (sure, on a different node but the architectural focus, power management ICs is what counts)? And Y-series stuff was crippled crap.

 

[Geddagod]

Intel knew that there was nothing in the world they could do to prevent Apple from dropping x86 Macs. Apple would have switched even if Intel was coming out with better CPUs than what Apple had, though Apple having something better or at least equivalent obviously smoothed the transition for them.

Intel knew that Apple's ability to steal x86 Windows market share was extremely limited, because they run a different OS, and they only compete on the high end not in the mass market. There would be little to gain by 'responding' to Apple because they cannot ever get Apple to make x86 Macs again and will be just as limited in their ability to steal Mac share as Apple is in its ability to steal Windows share. The equation changes if they believe Qualcomm will have a chance to make inroads into Windows market share (I'm still skeptical)

I think you would be surprised by how many people choose Macs not because of the "ecosystem" but more because of purely the battery life advantage. It's not something like an iPhone where there is pretty large amounts of social pressure to conform, take a nice x86 laptop outside and you will still get appreciation for having a nice machine lol.
At the very least, there's a much higher ability for Intel to try to steal back some of the lost Mac shares vs trying to convince Apple to switch back to Intel.

 

[FlameTail]

Lunar Lake will come fwiw, so Intel clearly already feels the heat. It’s also an expensive part on N3B though, so we’ll see how it goes.

Why does Lunar Lake use N3B??

 

[SpudLobby]

I think you would be surprised by how many people choose Macs not because of the "ecosystem" but more because of purely the battery life advantage.

Yeah, I think Doug is showing his age, social circles maybe with the assumption that Apple Silicon hasn’t courted some users significantly for the performance + battery life pairing they offer. They’re a hit with [American especially] developers and creatives in a way that puts their previous hold on said groups to shame. Often reluctant, but it is what it is.

AS Macs are full computers with AMD/Intel class performance or better below a point, and the efficiency/mixed use runtime of what feels like a scaled up iPad/Phone, which is a huge feat to everyone who keeps their neck trimmed — people here can snark all they want but it’s true.

And those users count given the portion of the “market share” they make up (e.g. they are not buying Chromebooks, they are affluent, or professionals, etc).

It's not something like an iPhone where there is pretty large amounts of social pressure to conform, take a nice x86 laptop outside and you will still get appreciation for having a nice machine lol.

Man. Exactly this. It really is about the hardware.
It’s not like iMessage in the US, this is not a social status play or a vehicle for communication at least to this demographic we’re referring to. Sure there are the Apple diehards for whom that is the purpose but there is a significant contingent of switchers that just want good hardware. (Also I agree that Mac vs PC is just shallow memes these days, people know a nice PC, it’s not frowned upon).


What this also means is if Intel et. Al had it together — or if QC gets it together just as likely — there are people that would be fine using a Windows laptop and they do not face the uphill battle that like Samsung and Google do in the US lol.

Alternatively I do think if you don’t see real competition on this front soon then Apple will gain more market share and from the demographics that matter if you care about the whole “turning a profit” thing. Not as bad as Android vs iOS margin shares of phones but things could still shift.

At the very least, there's a much higher ability for Intel to try to steal back some of the lost Mac shares vs trying to convince Apple to switch back to Intel.

Yep agree. They won’t switch back but there are definitely Mac users that would switch over if Intel/AMD/QC /Nvidia could build something that takes it to Apple and/or has a better pricing tier

 

[jpiniero]

I'd say most of the appeal is OSX (and/or that it's Not Windows). There may be some peer pressure, yes...

I do however think there are enough OEMs that are jealous of the MBP's battery life, and that's who Qualcomm will be targeting. But will people buy it?

 

[Doug S]

Intel’s Lunar Lake is nakedly an acknowledgment and response on the note of ultramobile from a company that has consistently failed to take mobile — even what they have left to defend — as seriously as it should beyond god awful LKF experiments and throwing wattmaxed blobs that never belonged in mainstream laptops.

Apple in the end obviously would use their own I agree.

Anyway, it’s pathetic Intel’s “we’re kinda taking this really cereal this time” LNL won’t be here until late 2024 early 2025. Why wasn’t that there in 2020 in principle (sure, on a different node but the architectural focus, power management ICs is what counts)? And Y-series stuff was crippled crap.

As we haven't seen Lunar Lake yet it is hard to claim it is Intel acknowledging and responding to anything. Whether Lunar Lake is a success or not as far as how well it compares to the competition power/performance wise, will depend far more on how Intel's fab technology than Intel's CPU designers. If they took their current designs and redid them for 18A I'm sure they'd look great in comparison too.

 

[DrMrLordX]

Why does Lunar Lake use N3B??

That would be really weird if they did.

 

[SpudLobby]

Why does Lunar Lake use N3B??

well what else are they going to use?

That would be really weird if they did.

It does. They needed something ready in time and I suppose as it stands decided N3B was the best bet between predictable and performant (or rather, efficient especially in this case.)

I am looking forward to seeing ARL 20A SKUs vs N3 SKUs still.

 

[FlameTail]

well what else are they going to use?

N3E/N3P?
Or why not their own nodes like 20A/18A ?

It does. They needed something ready in time and I suppose as it stands decided N3B was the best bet between predictable and performant (or rather, efficient especially in this case.)

I am looking forward to seeing ARL 20A SKUs vs N3 SKUs still.

Hmm

 

[SpudLobby]

As we haven't seen Lunar Lake yet it is hard to claim it is Intel acknowledging and responding to anything.

Lol.
https://www.pcworld.com/article/834...-will-be-a-specialized-low-power-pc-chip.html (2022)

Intel Lunar Lake CPU sample spotted in SiSoftware database - VideoCardz.com

Intel Lunar Lake enters testing phase Intel has begun testing its next-gen client CPU series known as Lunar Lake. The company has made it explicitly clear that Lunar Lake is not intended to serve as a high-end gaming architecture. Instead, it is slated for launch for low-power devices, and is...

videocardz.com

(It’s not 20 cores fwiw, it’s 8: and 4 Lion Coves at 3.9GHz and the other 4 Skymont cores at 2.6GHz)

https://x.com/iancutress/status/1613609998386331673?s=46 (2023 direct statement from Intel about a deliberate re-focus on perf/watt from an architecture standpoint, though in practice might mean more uncore and all)



Oh man. What could Intel be thinking with a 17W 4+4 part running modest clocks with a loud shouting from Gelsinger and other executives about “ultramobile” and “performance per watt” and “architecture from the ground up” (more important than the fabs) that just so happens to be concurrent with this design and a departure from frequency insanity, a monolithic device + a leading node. They’re quite vocal about LNL. And man if those clocks and “ultramobile” mentions or emphasis on architecture don’t point to Intel belatedly seeing what’s coming and what the average ultraportable consumer values — and that there’s a gap vs today’s laptops from Intel (or AMD) — I don’t know what does!

They’re tacitly acknowledging their arch just sucks for this stuff today. It’s an important product and it’s kind should’ve been taken seriously years ago.
(And before you compare to Lakefield, Intel was *never* this vocal about that garbage beyond token packaging firsts.)

Again the architecture is honestly as/more important than their fabs for this product.

Intel: Lunar Lakes Tapes Out, Meteor Lake on Track for 2023 Ramp

Meteor Lake to ramp in 2023, Lunar Lake ready for production in 2024.

www.tomshardware.com

Above is this summer and below is the quote from Pat.

“With MTL progressing well, it is now appropriate to look forward to Lunar Lake, which is on track for production readiness in 2024, having taped-out its first silicon,” said the head of Intel. “Lunar Lake is optimized for ultra-low power performance, which will enable more of our PC partners to create ultra-thin and light systems for mobile users.”

Whether Lunar Lake is a success or not as far as how well it compares to the competition power/performance wise, will depend far more on how Intel's fab technology than Intel's CPU designers. If they took their current designs and redid them for 18A I'm sure they'd look great in comparison too.

Lunar Lake is probably going to be on N3. It’s rumor known. It even taped out earlier this year as Pat said. It ain’t 18A based on that timeline. Possible it’s 20A but I really really doubt it.

Come on man.

 

[DrMrLordX]

well what else are they going to use?

N3E. Unless for some reason N3E isn't ready by then.

 

[SpudLobby]

N3E. Unless for some reason N3E isn't ready by then.

They designed it for N3B though is the issue afaict

 

[DrMrLordX]

They designed it for N3B though is the issue afaict

Pretty sure they'll port it over to N3E despite the expense.

 

[FlameTail]

LPDDR4: 3200 Mbps
LPDDR4X: 4233 Mbps
LPDDR5: 6400 Mbps
LPDDR5X: 8533 Mbps
LPDDR6: 12800 Mbps

So if the pattern holds, LPDDR6 will have 12800 Mbps speeds, which looks insane from where we are now. For reference, that means 200 GB/s bandwidth on a 128bit bus.

 

[gdansk]

MediaTek is already supporting "LPDDR5T" from Hynix at 9600 Mbps as an option for D9300.
So 12800Mbps is not even that insane from where we are now.

 

[SpudLobby]

Pretty sure they'll port it over to N3E despite the expense.

We went over that in another thread. Others think N3B but depends on their timeliness and the PDK. You don’t “port it” if it hasn’t even been released, that would just be a double tape out. Still possible but all depends on their timelines of development and how much they wanted N3 capacity secured/didn’t want to wait on any uncertainty with N3E, which might have been a good case for them staying safer with N3B even if it cost more.

I’ll partially punt on this since it’s not my angle here, it doesn’t matter. Everyone agrees it’s N3 derivative.

 

[Glo.]

LPDDR4: 3200 Mbps
LPDDR4X: 4233 Mbps
LPDDR5: 6400 Mbps
LPDDR5X: 8533 Mbps
LPDDR6: 12800 Mbps

So if the pattern holds, LPDDR6 will have 12800 Mbps speeds, which looks insane from where we are now. For reference, that means 200 GB/s bandwidth on a 128bit bus.

IF the pattern holds, LPDDR7 will offer speeds up to 25600 Mbps.

And that will offer 400 GB/s on 128 bit bus.

Signal integrity will be a challenge already with LPDDR6 and 12800 MHz, let alone twice that frequency.

 

[igor_kavinski]

Lunar Lake is probably going to be on N3. It’s rumor known. It even taped out earlier this year as Pat said. It ain’t 18A based on that timeline. Possible it’s 20A but I really really doubt it.

Why doubt it? It could be early 20A silicon with really bad yields. Maybe that was the only working CPU they got out of the wafer that they then used in the demo laptop to show off.

 

[DrMrLordX]

I’ll partially punt on this since it’s not my angle here, it doesn’t matter. Everyone agrees it’s N3 derivative.

We'll see when we get there. I guess. Though it may take awhile to get solid information on the chips even after they're released.

 

[FlameTail]

Hey guys.

Some thoughts...



Oryon: 3227@4.3 GHz
M2 Max: 2841@3.7 GHz

They say Oryon can match M2 Max performance at 30% less power. That's significant.

So for Oryon to match M2 Max... it's clockspeed would be:

3.78 GHz.

Which is roughly equal to the base clock speed of 3.8 GHz.

_______

Oryon at 3.8 GHz matches M2 Max ST performance at 30% less power.

That is significant. The performance per watt of this core looks great. I guess this is what Amon and Gerard were referring to when they talked about how the real silicon turned out to be much better than the simulations. It's not the raw performance, but the efficiency.

But that isn't my point.

My point is... why did they then limit the base clock speed to 3.8 GHz?

Why the need for Dual-core boost?

Considering how good the performance per watt is, they could run all cores at 4.3 GHz. Or atleast 4 GHz like Apple's doing in the M3 series.

Unless... as the rumours have us suspecting- there are deficiencies in the uncore.

That would explain why the all-core speed is limited to 3.8 GHz, and also why the X Elite- a CPU with 12 P cores is underperforming in MT tests.

 

[Thibsie]

Don't forget this is a server core. The power aspect isn't the first one when developing such a core. I don't think they used all possibilities to power gate areas as they would have done on a phone SOC.

With proper time and implementation, they're rather low hanging fruits to be taken IMO and power is one.

 

[FlameTail]

MediaTek is already supporting "LPDDR5T" from Hynix at 9600 Mbps as an option for D9300.
So 12800Mbps is not even that insane from where we are now.

Of course, Hynix has LPDDR5T-9600 and even Micron has 9600, although they are calling it just LPDDR5X-9600.

I wonder if future APUs/SoCs from AMD and Qualcomm will use these? Are these 9600 Mbps RAM produced in high volume? My impression is that it will only be used in niche high-end Android smartphones.

Something like Strix Point/Halo could benefit from LPDDR-9600, which would be substantial uplift from the LPDDR5X-7500 they currently support.

 

[FlameTail]

MediaTek is already supporting "LPDDR5T" from Hynix at 9600 Mbps as an option for D9300.
So 12800Mbps is not even that insane from where we are now.

Of course, Hynix has LPDDR5T-9600 and even Micron has 9600, although they are calling it just LPDDR5X-9600.

I wonder if future APUs/SoCs from AMD and Qualcomm will use these? Are these 9600 Mbps RAM produced in high volume? My impression is that it will only be used in niche high-end Android smartphones.

Something like Strix Point/Halo could benefit from LPDDR-9600, which would be substantial uplift from the LPDDR5X-7500 Phoenix supports.

 

[FlameTail]

Don't forget this is a server core.

Is it though?

The Phoenix core for servers was developed by Nuvia before they got acquired hy Qualcomm. My impression is that the Oryon core is not what Phoenix was, judging from the "Ian Cutress x Gerard Williams" interview.

Also didn't ARM order Qualcomm to destroy all pre-acquisiiton Nuvia IP as part of the whole lawsuit fiasco?

 

[FlameTail]

Do you guys reckon that the Oryon CPU has custom ARM ISA extensions like Apple Silicon does?