Discussion Qualcomm Snapdragon Thread

[SteinFG]

Is Purwa not single channel memory after all?

ok, so they weren't so dumb to leave free performance on the table. good.

 

[FlameTail]

Notebookcheck has analysed the 8-core X Plus!!!

Qualcomm Snapdragon X Plus 8-core analysis - The X1P-42-100 is smaller and cheaper, but not very efficient

Notebookcheck analysis of the Snapdragon X Plus X1P-42-100 with 8 CPU cores and the slower X1-45 Adreno GPU.

www.notebookcheck.net

 

[GTracing]

Notebookcheck has analysed the 8-core X Plus!!!

Qualcomm Snapdragon X Plus 8-core analysis - The X1P-42-100 is smaller and cheaper, but not very efficient

Notebookcheck analysis of the Snapdragon X Plus X1P-42-100 with 8 CPU cores and the slower X1-45 Adreno GPU.

www.notebookcheck.net

It seems a bit misleading to use multi core benchmarks to compare "efficiency" between laptop chips. It makes the 12 core X Elite SKUs look more "efficient" compared to the lower end SKUs. In reality, 95% of the workloads these chips will be subjected to won't use 8+ cores. Comparing power consumption with web browsing, Microsoft office, video playback, etc. would be a much more useful comparison. Good article nonetheless.

 

[DZero]

Currently released SKUs:

X1E-00-1DE
X1E-84-100
X1E-80-100
X1E-78-100

X1P-66-100
X1P-64-100

X1P-46-100
X1P-42-100

Rumoured SKUs;

X1E-76-100

X1P-62-100
X1P-56-100
X1P-44-100
X1P-40-100
X1P-39-100

X1-24-100
X1-00-001

Mmm.. I will highlight the potential chips to be released.

I can see the X1P-56 variant being released as an underclocked variant of the X1P, the rest are so near.
And the X1-24 being the most basic processsor.

EDIT: X1-00 is the Dev kit. So, X1-24 will be the bottom.

 

[FlameTail]

Notebookcheck has analysed the 8-core X Plus!!!

Qualcomm Snapdragon X Plus 8-core analysis - The X1P-42-100 is smaller and cheaper, but not very efficient

Notebookcheck analysis of the Snapdragon X Plus X1P-42-100 with 8 CPU cores and the slower X1-45 Adreno GPU.

www.notebookcheck.net

Also they seem to have found that the X1P-42-100/Purwa has only performance clusters.

Hamoa is
2× Performance clusters
1× Efficency cluster

So I thought Purwa would be;
1× Performance cluster
1× Efficiency cluster

But if what Notebookcheck is saying is true, then Purwa is;
2× Performance clusters
No efficency clusters

That is supremely bizarre, even from an engineering standpoint.

I am quite skeptical about this until someone else's findings align with it.

There is also the matter of whether Purwa has a 64 bit or 128 bit memory bus, which I am still not certain about.

 

[Ghostsonplanets]

As I said, pricing will be key. I haven't seen a single laptop announced that meet Cristiano promised target of $700.

And there is the X1-24-100 rumours... makes me think how slow it will be.

This supposedly exists, yeah. I wonder if it will be a 6-core binned SKU from Purwa

 

[FlameTail]

And there is the X1-24-100 rumours... makes me think how slow it will be.

Place your bets.

It will be bronze coloured.

X Elite = Gold
X Plus = Silver
X = Bronze

 

[ikjadoon]

The naming remains a hot mess. How can some X Plus SKUs have higher 1T perf than an X Elite SKU? Is Oryon also meant to "disrupt" the industry-standard that in a stack, higher-end CPUs mean higher nT and 1T perf? I reiterate my belief that these terrible names are an attempt to mislead people into buying a much-worse performing Oryon SKU.

//

Always excluding GB6 1T when it's convenient, sigh. On a lower-end chip, nT likely means even less.

 

[FlameTail]

I reiterate my belief that these terrible names are an attempt to mislead people into buying a much-worse performing Oryon SKU.

How does that help Qualcomm though?

The X1E-78-100 has been quite controversial, and for good reason. The lack of Core Boost means it's a major step down in 1T performance. Unlike nT performance which is limited by TDP and thermals, 1T performance actually matters for system responsiveness, web browsing and general usage.

Some have pointed out how it's hypocritical that Qualcomm uses the top end X1E-84 to compare with Intel and AMD, while most X Elite laptops actually ship with the X1E-78.

I'd say the best thing Qualcomm could do is eventually phase out the production of the X1E-78.

 

[FlameTail]

Acer Go 14 AI announced with Snapdragon X;

Acer’s 14-inch laptops claim 24 hours of battery life from Intel, Qualcomm, or AMD

Your pick.

www.theverge.com

They are saying Snapdragon X desktops are coming...

Qualcomm CEO confirms Snapdragon desktops are planned

According to Qualcomm chief executive Cristiano Amon, a high-end Snapdragon desktop chip is in the cards.

www.pcworld.com

Where's the dev kit, anyways?

 

[DZero]

Place your bets.

It will be bronze coloured.

X Elite = Gold
X Plus = Silver
X = Bronze

Indeed, but I want to know the core configuration. I expect 4+4 but at lowest clock.

Also they seem to have found that the X1P-42-100/Purwa has only performance clusters.

Hamoa is
2× Performance clusters
1× Efficency cluster

So I thought Purwa would be;
1× Performance cluster
1× Efficiency cluster

But if what Notebookcheck is saying is true, then Purwa is;
2× Performance clusters
No efficency clusters

That is supremely bizarre, even from an engineering standpoint.

I am quite skeptical about this until someone else's findings align with it.

There is also the matter of whether Purwa has a 64 bit or 128 bit memory bus, which I am still not certain about.

Unless the 2X performance clusters are aimed to Desktop, could have more sense.

 

[igor_kavinski]

Where's the dev kit, anyways?

They are overwhelmed by demand and currently torturing chips to see if they can make the cut.

 

[adroc_thurston]

The naming remains a hot mess.

It's not a naming issue, it's a binning issue.
They have real chungus spreads on bins per die, which is why top-end Purwa bin runs a lot faster than a shoddy binned Hamoa.

 

[FlameTail]

Oryon V1 as implemented in Snapdragon X1 seems to have a number of shortcomings;
• Apparently wide binning/yield variation if the rumours are to be believed
• It drinks a lot of power (compared to Apple CPUs for example. It's certainly not reaching the crazy efficiency levels Nuvia promised with Phoenix in 2021).

It will be interesting to see if the Snapdragon 8 Gen 4 fixes these issues. Rumours are sounding good:
• 8G4 regular version has a peak clock of 4.37 GHz. Qualcomm is going to be shipping tens of millions of units of this chip. That's a sign that they have got the yields/binning under control.
• 8G4 was rumoured to hit 9.5W in single core power consumption. While this sounds quite alarming for a phone SoC, if you compare it to X Elite, it's a great improvement.

8G4 = 4.37 GHz = 9.5W
X Elite = 4.2 GHz = 15W+

 

[ikjadoon]

How does that help Qualcomm though?

...
Some have pointed out how it's hypocritical that Qualcomm uses the top end X1E-84 to compare with Intel and AMD, while most X Elite laptops actually ship with the X1E-78.

I'd say the best thing Qualcomm could do is eventually phase out the production of the X1E-78.

I presume that consumers read a great X Elite review of a chart-topping SoC (e.g., The Verge), search for a product, see and rapidly ignore the confusing SKU name ("Hey, it's X Elite, like the review!"), and then purchase a slower device that cost Qualcomm less (e.g., binning) than what they expected.

It's a little shrewd since the gap is wide, but not as problematic as some other schemes by AMD & Intel because the base CPU is still fast for most consumers (e.g., ~2500 Pts in GB6, which is equivalent to an i7-12700). It may not be as fast as they read in reviews, but it won't be like accidentally buying a Celeron PC.

Unless you accidentally buy a Snapdragon 8cx Gen3 or whatever, so Qualcomm also has a pretty low perf floor.

Maybe the lower-end chips, like the 78, suffer more noticeably under emulation than the 80 chips, but I have nothing but speculation to stand behind that claim.

 

[FlameTail]

Samsung has announced a new Galaxy Book4 Edge 15" with the 8-core X Plus. It seems to be lower tier than the 14"/16" versions.

Galaxy Book4 Edge, Samsung’s Next-Gen AI PC, Now Available With 15-Inch Display

All the power of Copilot+ PCs, complete with access to Galaxy AI features through the connected Galaxy ecosystem, in a compact and lightweight design

news.samsung.com

So following the announcement of the 8-core X Plus by Qualcomm, we are also seeing a flurry of device announcements with it from OEMs.

 

[poke01]

8G4 = 4.37 GHz = 9.5W

Isn’t that like higher than M4??

My guess is only gaming phones will achieve that.

 

[Ghostsonplanets]

Acer Go 14 AI announced with Snapdragon X;

Acer’s 14-inch laptops claim 24 hours of battery life from Intel, Qualcomm, or AMD

Your pick.

www.theverge.com

They are saying Snapdragon X desktops are coming...

Qualcomm CEO confirms Snapdragon desktops are planned

According to Qualcomm chief executive Cristiano Amon, a high-end Snapdragon desktop chip is in the cards.

www.pcworld.com

Where's the dev kit, anyways?

$1199 is this a joke? QCOM won't get any meaningful marketshare with such pricing.

 

[ikjadoon]

• 8G4 was rumoured to hit 9.5W in single core power consumption. While this sounds quite alarming for a phone SoC, if you compare it to X Elite, it's a great improvement.

8G4 = 4.37 GHz = 9.5W
X Elite = 4.2 GHz = 15W+

FWIW, I imagine like Arm's X925 vs X4 disclosures, I believe these leaks also include TSMC N4(x) → N3E improvements, i.e., I don't think any of the upcoming mobile SoCs use an N4-class node.

Not big changes, but some improvements nonetheless.

 

[FlameTail]

Ian Cutress gave an update about the Dev Kits during the Tech Poutine livestream.

Dev Kits will ship in November. Qualcomm supposedly had an issue with the HDMI ports not working properly. Also it seems Qualcomm underestimated the demand for the dev kit, so now they are engaging with their ODM to produce more of them

 

[DZero]

Dev kits at this pace will end totally useless and sell it as mini PCs.

 

[jdubs03]

Isn’t that like higher than M4??

My guess is only gaming phones will achieve that.

Yes by about 2W or so.

 

[FlameTail]

Isn’t that like higher than M4??

My guess is only gaming phones will achieve that.

Depends how the power is measured

 

[FlameTail]

Hamoa and Purwa motherboards;

Picture from SnazzyLabs

I wonder if anyone has tried measuring Purwa's die size?

 

[POWER4]

How does that help Qualcomm though?

The X1E-78-100 has been quite controversial, and for good reason. The lack of Core Boost means it's a major step down in 1T performance. Unlike nT performance which is limited by TDP and thermals, 1T performance actually matters for system responsiveness, web browsing and general usage.

Some have pointed out how it's hypocritical that Qualcomm uses the top end X1E-84 to compare with Intel and AMD, while most X Elite laptops actually ship with the X1E-78.

I'd say the best thing Qualcomm could do is eventually phase out the production of the X1E-78.

Agree 100%.

I presume that consumers read a great X Elite review of a chart-topping SoC (e.g., The Verge), search for a product, see and rapidly ignore the confusing SKU name ("Hey, it's X Elite, like the review!"), and then purchase a slower device that cost Qualcomm less (e.g., binning) than what they expected.

This is only true if reviewers were mostly testing units with X1E-80 and X1E-84. Since some claimed most come with X1E-78, I think this suspicion does not hold.

Also, I don't know how relevant it is to user experience. Sure, it should make a difference, but how much? What is the point of diminishing returns in terms of user experience? Being guided only by benchmark numbers is like walking blind. Call me a fanboy, but my experience with the Surface Laptop, which IIRC uses the X1E-78, was enough to leave me impressed.