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

 

[Doug S]

2600 points is below the year old A16 and won't be arriving in phones until over a year from now. How is it even with some of Apple's top designers and Apple showing meager gains the last few iterations Qualcomm will still be over two years behind?

 

[jpiniero]

2600 points is below the year old A16

Average A16 score is in the ~2500 range, according to GB's site. If the power consumption is reasonable, they are catching up.

 

[soresu]

2600 points is below the year old A16 and won't be arriving in phones until over a year from now. How is it even with some of Apple's top designers and Apple showing meager gains the last few iterations Qualcomm will still be over two years behind?

They say it takes a village to raise a child, but it takes a village of very smart and hard working people years to make a competitive CPU µArch these days - and even then they are aiming at a target that may well be entirely inadequate at the time they put it on the drawing board.

Plus as Apple control the full software stack excepting the apps themselves they can pull some tricks in core design that neither Qualcomm or even ARM can do.

 

[igor_kavinski]

How is it even with some of Apple's top designers and Apple showing meager gains the last few iterations Qualcomm will still be over two years behind?

They are clearly enjoying their time at Qualcomm and working at a leisurely pace and that must have been one of the clauses in their contract. What choice does Qualcomm have other than to treat them like VIPs? Qualcomm seems like a great place to work for people who can pretend to do more than they actually do. Remember Renduchintala and what he did for Intel? They were forced to fire his Ph.D butt for being ineffective.

 

[Thibsie]

Qualcomm has an enterprise culture problem for what I understood. A big one. And egos in the way too.

 

[Doug S]

Average A16 score is in the ~2500 range, according to GB's site. If the power consumption is reasonable, they are catching up.

Those scores are with GB 6.0, with 6.2 the scores went up (and safe to say any "leaked" benchmarks will use whatever gives the highest number so they will be 6.2 results)

Why Geekbench adds to the confusion with crap like that I don't know, but scores for A16 on 6.2 are within spitting distance of 2700 (remember A17P is only a 10% gain and got 3000...if A16 was 2500 that would mean A17P was 20% better and no one would be talking about Apple's small gains the last few generations)

 

[hemedans]

Those scores are with GB 6.0, with 6.2 the scores went up (and safe to say any "leaked" benchmarks will use whatever gives the highest number so they will be 6.2 results)

Why Geekbench adds to the confusion with crap like that I don't know, but scores for A16 on 6.2 are within spitting distance of 2700 (remember A17P is only a 10% gain and got 3000...if A16 was 2500 that would mean A17P was 20% better and no one would be talking about Apple's small gains the last few generations)

What about Clock speed? They didn't mention at what clock speed Nuvia score 2600, that's 8 gen 4 score it's for phones it may have much lower clocks peed, not every one will clock their phone soc to 3.7Ghz. There is possibility for 8 gen 4 to have more IPC than A16/A17.

 

[Nothingness]

Those scores are with GB 6.0, with 6.2 the scores went up (and safe to say any "leaked" benchmarks will use whatever gives the highest number so they will be 6.2 results)

Why Geekbench adds to the confusion with crap like that I don't know, but scores for A16 on 6.2 are within spitting distance of 2700 (remember A17P is only a 10% gain and got 3000...if A16 was 2500 that would mean A17P was 20% better and no one would be talking about Apple's small gains the last few generations)

As far as I can see A16-based iPhone's are closer to 2600 than to 2700 on Geekbench 6.2.
But score searching is such a mess, I might be wrong. I wonder if I won't download the xx M v6 results locally and build a DB

 

[DrMrLordX]

As far as I can see A16-based iPhone's are closer to 2600 than to 2700 on Geekbench 6.2.
But score searching is such a mess, I might be wrong. I wonder if I won't download the xx M v6 results locally and build a DB

To further complicate matters, Apple is releasing/has released an iOS update to deal with overheating issues on the iPhone 15:

Apple will issue a software update to address iPhone 15 overheating complaints

Apple said that the new iPhones were running hot because of a combination of bugs in iOS 17, bugs in apps, and a temporary set-up period.

www.cnbc.com

Reading between the lines, it looks like A16 is too much for the iPhone's cooling capabilities. Apple likely let it run extra hot to hit performance targets (or get as close as they could). Now that they've flooded the market with benchmarks of the release firmware, they can throttle it back and pretend nothing's wrong.

It shouldn't surprise anyone if A16 hit lower GB6.2 scores after the update.

 

[Thibsie]

Yeah, it's called cheating IMHO.

 

[Saylick]

To further complicate matters, Apple is releasing/has released an iOS update to deal with overheating issues on the iPhone 15:

Apple will issue a software update to address iPhone 15 overheating complaints

Apple said that the new iPhones were running hot because of a combination of bugs in iOS 17, bugs in apps, and a temporary set-up period.

www.cnbc.com

Reading between the lines, it looks like A16 is too much for the iPhone's cooling capabilities. Apple likely let it run extra hot to hit performance targets (or get as close as they could). Now that they've flooded the market with benchmarks of the release firmware, they can throttle it back and pretend nothing's wrong.

It shouldn't surprise anyone if A16 hit lower GB6.2 scores after the update.

Fwiw, apparently iOS 17.0.3 mitigates higher temps without compromising performance:

https://twitter.com/x/status/1710638738693636423

 

[igor_kavinski]

Fwiw, apparently iOS 17.0.3 mitigates higher temps without compromising performance:

https://twitter.com/x/status/1710638738693636423

Sums it up nicely.

 

[Doug S]

Fwiw, apparently iOS 17.0.3 mitigates higher temps without compromising performance:

That fits with the evidence that it was GPU driver immaturity that was at least partially if not mostly responsible for the issues. Not terribly unexpected with a brand new GPU. There is a 17.0.4 coming out that's supposed to address some other issues, but if there are still some overheating scenarios left it may be intended to fix that also.

The thing that made me really believe it was GPU related was a video showing iPhone 15 Pro overheating while sitting static in Instagram. Clearly that's not something that could overtax the CPU cores, but it is easy to see scenarios that could overtax the GPU if a buggy driver interacting with certain apps was making it do something stupid like rendering a static screen 120 times a second despite it not changing, or prerendering everything Instagram was downloading in the background (to support doomscrolling) without limit.

 

[FlameTail]

https://x.com/Tech_Reve/status/1711529393766695222?s=20

MOAR CACHE!

 

[DrMrLordX]

Fwiw, apparently iOS 17.0.3 mitigates higher temps without compromising performance:

Drat, cynicism loses. This time.

Honestly though I'm surprised they were able to pull that off.

 

[FlameTail]

So will this iOS update affect the power curves?

I dunno what exactly the update did, but let's say if Geekerwan tests the the A17 Pro again...

 

[igor_kavinski]

Honestly though I'm surprised they were able to pull that off.

I'll be surprised if they can squeeze more performance out of it just ONE more time with another update. Seems TSMC is up against a wall so it is in Apple's best interests to start prototyping their chips on Intel's (and maybe Samsung's) current and future test nodes to prevent performance stagnation in case their arch team has no more tricks left to get more out of their ARM architecture. They may need to depend heavily on manufacturing node performance efficiency from now on.

 

[Doug S]

I'll be surprised if they can squeeze more performance out of it just ONE more time with another update. Seems TSMC is up against a wall so it is in Apple's best interests to start prototyping their chips on Intel's (and maybe Samsung's) current and future test nodes to prevent performance stagnation in case their arch team has no more tricks left to get more out of their ARM architecture. They may need to depend heavily on manufacturing node performance efficiency from now on.

Just look at what Nvidia and AMD typically do with a brand new GPU architecture, and how it gains performance over time as the drivers become more mature. I see no reason the same thing shouldn't be true for Apple. They certainly won't have any magic ability to write quality drivers for new GPU hardware from date of release if the kingpins of GPUs can't.

It never made sense to me that Apple has a brand new GPU, went from 5 to 6 cores, and were claiming only 20% performance gain. Why would you even have a new GPU architecture if it performed exactly the same as the old one and your only performance increase came from increasing the number of cores by 20%?

They probably knew the drivers were immature and set a level they knew they could reach, rather than make claims based on where they think the GPU will be once they have the bugs shaken out and the drivers properly tuned.

 

[DrMrLordX]

I'll be surprised if they can squeeze more performance out of it just ONE more time with another update. Seems TSMC is up against a wall so it is in Apple's best interests to start prototyping their chips on Intel's (and maybe Samsung's) current and future test nodes to prevent performance stagnation in case their arch team has no more tricks left to get more out of their ARM architecture. They may need to depend heavily on manufacturing node performance efficiency from now on.

You may be giving 18a too much credit.

 

[Frenetic Pony]

To further complicate matters, Apple is releasing/has released an iOS update to deal with overheating issues on the iPhone 15:

Apple will issue a software update to address iPhone 15 overheating complaints

Apple said that the new iPhones were running hot because of a combination of bugs in iOS 17, bugs in apps, and a temporary set-up period.

www.cnbc.com

Reading between the lines, it looks like A16 is too much for the iPhone's cooling capabilities. Apple likely let it run extra hot to hit performance targets (or get as close as they could). Now that they've flooded the market with benchmarks of the release firmware, they can throttle it back and pretend nothing's wrong.

It shouldn't surprise anyone if A16 hit lower GB6.2 scores after the update.

It's just loop cycling some routine by accident until the whole thing overheats, looks like some interaction that worked fine on all previous ios versions but the new version changes the behavior blah blah blah classic software overheat bug.

Pretty much all phone manufacturers run their phone at 5w initially until heat buildup requires them to lower it to about 3.5w. It's not even cheating so much as the expectation that initial tasks will be highly bursty, so executing them as fast as possible is a priority and running into heat issues isn't expected because the activity won't last long. But a lot of benchmarks can run at max speed for a long time, thus throttling sets in there even if it wouldn't for burst activity.

It gets worse as well. We can look back to the Samsung/AMD 920 vs Snapdragon, uhm, stupidity, to see why manufacturers shouldn't throttle immediately. The Exynos with the custom AMD/Samsung GPU in it was in a phone that was set to throttle for long term stability. The Qualcomm chip was in otherwise the same phone, and was not set to throttle. Headlines screamed that the Qualcomm one was faster and the custom GPU "an utter failure", a narrative that survives to this day, despite different testing conditions. When set to the same conditions it turned out the custom AMD GPU was slightly better in terms of performance per watt than the Qualcomm one.

So not only is throttling not cheating, at least when it comes to quick bursts, it also needs to be off by default because the internet will be filled with screaming headlines about how much of a useless failure the phone/chip/etc. is if throttling is on, even throttling is a better benchmark for gaming (which is obviously expected to last long enough to throttle). Phone manufacturers don't really have a choice here.

 

[Saylick]

Looks like we get a new name:

Introducing an all-new naming convention for our next-generation of intelligent PC platforms: The Snapdragon X Series

Snapdragon X compute platforms will deliver next-level performance, AI, connectivity and battery life, building on our years of experience engineering heterogeneous compute architectures across the CPU, GPU and NPU.

www.qualcomm.com

 

[Orfosaurio]

So will this iOS update affect the power curves?

I dunno what exactly the update did, but let's say if Geekerwan tests the the A17 Pro again...

I suspect that he doesn't want to do it, he have shown a bias against Apple, or maybe towards SoCs that chinese manufactures can use. At least he is no were as close as being as biased like GoldenReviewer.

 

[ikjadoon]

2600 points is below the year old A16 and won't be arriving in phones until over a year from now. How is it even with some of Apple's top designers and Apple showing meager gains the last few iterations Qualcomm will still be over two years behind?

https://x.com/Tech_Reve/status/1710684439595057411?s=20

Nuvia core in Snapdragon 8 gen 4 reportedly reaching 2600-2900 points.

To be fair, the Snapdragon 8 Gen 2 for Galaxy scores ~1960 in GB6.0 and the X4 is not much faster.

Cortex-X3 @ 3.7 GHz: 1964 GB6.0
Cortex-X4 @ 3.7 GHz: 2101 GB6.0 (Arm estimate of ~1.07x uplift)

But, the 8G4 (2025) will need to compete with the yet-unannounced Cortex-X5 from Arm in 2025 and Apple's "A18" SoC in late 2024.

I'm assuming it's GB6.0, but Arm isn't specific.

 

[moinmoin]

Looks like we get a new name:

Introducing an all-new naming convention for our next-generation of intelligent PC platforms: The Snapdragon X Series

Snapdragon X compute platforms will deliver next-level performance, AI, connectivity and battery life, building on our years of experience engineering heterogeneous compute architectures across the CPU, GPU and NPU.

www.qualcomm.com

The Snapdragon "formerly Twitter" series?

(The usage of "X" really leaves a bad taste to me these days.)

 

[FlameTail]

The Snapdragon "formerly Twitter" series?

(The usage of "X" really leaves a bad taste to me these days.)

My first impressions of reading that was positive.

*A clear, simplified tiering structure helps users navigate our platform capabilities from mainstream to premium.*

Seems they are committed to having a proper and clear naming scheme. Much appreciated coming the from their "8cx Gen 3" or the whack Intel unveiled recently.