Discussion Qualcomm Snapdragon Thread

[adroc_thurston]

How do they plan on competing with STX1 then

They don't?
No one is.

I thought the entire shtick was that it would be more efficient at comparable perf.

Should still win some on battery life! hopefully.
Rest yeah, Hamoa is bad Z5c basically.
Which I've said like a zillion times already. Maybe not here.

 

[poke01]

Oh god, phone companies making real hardware. Pain.

It’s not even a phone company. It’s a modem company. They can’t even make a laptop SoC launch on time. Wonder where all the experience with WoA went since 2016.

 

[Hitman928]

Is that SoC package power or any chance it is actually platform power as QC has quoted so far?

TBH, their power numbers were always very suspect to me and I indicated as much when they first released them but then I decided it was best to remain quiet as I had no evidence against them. If these are SoC power numbers, then their perf/W graphs they've shown are basically trash. If it's platform power, there's a little more wiggle room but could upgrade from trash to deceptive.

 

[adroc_thurston]

Wonder where all the experience with WoA went since 2016.

Oh that BU was a mess. Still is.

Is that SoC package power

this.

If these are SoC power numbers, then their perf/W graphs they've shown are basically trash

well yeah the reality of mass producing silicon hits.
Remember, fmax on the top bin went 100Mhz down.

 

[Kepler_L2]

OOOOF I guess it's a good thing Flametail took a break:


View attachment 97827

View attachment 97828

But I thought ARM was more efficient

 

[adroc_thurston]

But I thought ARM was more efficient

In generating cope yes.
Please post the juanrga.png for me.

 

[Kepler_L2]

In generating cope yes.
Please post the juanrga.png for me.

 

[ikjadoon]

View attachment 97828

Assuming this is true (Android Authority rarely has its own leaks):

It's rare / the first time I've ever seen power draw per bin released for individual CPU SKUs. Some of this variance is huge, so it makes sense why Qualcomm may have up to 8x SKUs.

Yet, confusingly, these are per-SKU, but then Android Authority says "In practice, the better parts will simply get binned as the higher-end SKUs, though, so the distinction doesn’t have any significant implications in reality." But these are already per SKU. There is no more segmentation, as far as we see.

So, one reviewer may get an X1E84100 that draws 98.5W, while another X1E84100 draws 82.3W? The former consumes +19% more power. That's not close, is it? Battery life tests between different units may be quite variable, especially under heavy loads (e.g., I assume these are maximum / near-peak power draw).

EDIT: and 1 out of 20 X1E84100 units may consume more than 98.5W? There's something funky here with these numbers (or this SoC?), but I'll admit I'm not familiar with this variance in other mobile SoCs.

//

But, meta aside, I am a bit sad the lower-end X1 Plus SKU by default can push up to 35W+. That range of 35W-43W will not fly on a fanless device. Hopefully OEMs will have cTDP / TDP-down configurations, but disappointed why Qualcomm didn't limit them to ~20W max. on at least 1-2 lower-end SKUs.

Windows OEMs are notorious for copying the CPU manufacturer's default TDP and designing the laptop around it. So users invariable will get the added weight for larger heatsinks, fans, inlet / outlet, dust maintenance, etc., even if there is a software-configurable "quiet / fanless mode".

 

[adroc_thurston]

It's rare / the first time I've ever seen power draw per bin released for individual CPU SKUs. Some of this variance is huge, so it makes sense why Qualcomm may have up to 8x SKUs.

Qualcomm, being a phone vendor, has no concept of power limits.
Just clock targets and tjmax.

So, one reviewer may get an X1E84100 that draws 98.5W, while another X1E84100 draws 82.3W? The former consumes +19% more power.

No different from Intel desktop chips in DIY mobos, which this is a mobile equivalent of.

 

[ikjadoon]

The article reporting is just as vague as Qualcomm's marketing slides, if not even more so. Charlie is essentially asking his audience to trust him.

I think @ikjadoon did a better job at finding reasons to doubt QC than Semiaccurate, and did so with publicly available information.

One qualification to note: I assume Qualcomm has consistently used Geekbench 6.1/6.2 for all publicly released numbers.

I forgot GB6.3 released two weeks ago and it increased some Arm CPU scores by adding support for Arm's SME instructions. For Arm systems with SME, GB scores aren't comparable according to John Poole (GB founder):

GB6.3 = not comparable to any other GB 6.x
GB 6.2 = GB6.1
GB 6.0 = not comparable to any other GB6.x

For systems without SME instructions, Geekbench 6.3 CPU Benchmark scores are comparable with Geekbench 6.1 and Geekbench 6.2 scores. Systems with SME instructions enabled will score higher in Geekbench 6.3 than in earlier Geekbench versions.. Geekbench 6.3 GPU Benchmark scores are compatible with Geekbench 6.2 for all systems.

//

But that may not matter, as IIRC, SME is only in Armv9.x designs and from the lawsuit, Qualcomm has hinted Oryon is (was?) Arm v8.x (Feb 24, 2024 document):

 

[Henry swagger]

https://wccftech.com/snapdragon-x-elite-package-power-can-reach-almost-100w/ reaches 100watts.. apple cores are still king in efficiency

 

[Nothingness]

I forgot GB6.3 released two weeks ago and it increased some Arm CPU scores by adding support for Arm's SME instructions. For Arm systems with SME, GB scores aren't comparable according to John Poole (GB founder):

GB6.3 = not comparable to any other GB 6.x
GB 6.2 = GB6.1
GB 6.0 = not comparable to any other GB6.x

Sorry but no, it increased no score as no available Arm CPU has SME.

 

[DrMrLordX]

juanrga

Oh no, he's back? We're doomed!

Nobody can stop the Internet Strongman!

 

[okoroezenwa]

https://wccftech.com/snapdragon-x-elite-package-power-can-reach-almost-100w/ reaches 100watts.. apple cores are still king in efficiency

Interestingly Andrei is over on Reddit disputing these power usage numbers.

 

[Henry swagger]

Interestingly Andrei is over on Reddit disputing these power usage numbers.

What did he say ?

 

[okoroezenwa]

What did he say ?

It’s linked to “Reddit” in my comment (is it not showing up?) but he says the numbers aren’t accurate and we should wait for launch products ¯\_(ツ)_/¯

 

[Hitman928]

What did he say ?

He doesn’t say it’s entirely wrong but that the info is misinterpreted and the portrayal is wrong. Not sure what that means though. Seems like the power spread is legit, but the SKU breakdown isn’t correct, something like that.

The table is misinterpreted and wrong as how it's portrayed - it's not per-SKU power variance, you should just wait for actual products. The workload is also not something realistic.

Edit: He also implies that the workload to get those numbers is something like Prime95 which pushes the chip further than any normal program. This may be true but things like Blender can also push chips pretty hard and may not be too far off these numbers either. Not too long before release and hopefully we get some good independent reviews to go along with it.

 

[Nothingness]

Not too long before release and hopefully we get some good independent reviews to go along with it.

Definitely. At the moment, it looks like various people are playing a clickbait game. I don't blame them, but I'll wait for public reviews.

 

[SarahKerrigan]

Definitely. At the moment, it looks like various people are playing a clickbait game. I don't blame them, but I'll wait for public reviews.

Yep. We'll know in weeks.

I don't think we'd be seeing the level of OEM interest that we are if the situation were as bad as some here seem to believe, though. Certainly OEMs were not lining up to ship premium Kaveri or Richland devices. I also freely admit that I have no experience directly dealing with the OEM side of the industry, so maybe I'm way off base.

 

[SpudLobby]

Interestingly Andrei is over on Reddit disputing these power usage numbers.

Yep. He also said in another forum to basically just ignore them.

 

[SpudLobby]

People have to stop being so neurotic.

Even Qualcomm’s own graphs still have them ahead of AMD/Intel on ST efficiency and with a much lower power floor. I expect this to be obvious in reviews wrt battery life, albeit less so vs LNL or Strix.

 

[Gideon]

Yep. He also said in another forum to basically just ignore them.

Which forum?

 

[adroc_thurston]

Even Qualcomm’s own graphs still have them ahead of AMD/Intel on ST efficiency and with a much lower power floor. I expect this to be obvious in reviews wrt battery life, albeit less so vs LNL or Strix.

Ugh yeah about that.

 

[SpudLobby]

Which forum?

Chips n Cheese discord. You’ll see him also say adroc is a liar etc.

 

[SpudLobby]

He doesn’t say it’s entirely wrong but that the info is misinterpreted and the portrayal is wrong. Not sure what that means though. Seems like the power spread is legit, but the SKU breakdown isn’t correct, something like that.



Edit: He also implies that the workload to get those numbers is something like Prime95 which pushes the chip further than any normal program. This may be true but things like Blender can also push chips pretty hard and may not be too far off these numbers either.

Not too long before release and hopefully we get some good independent reviews to go along with it.

The real takeaway here anyways