Discussion Qualcomm Snapdragon Thread

[adroc_thurston]

That's the idea. Though 2+4 is ehh. Poverty can't complain much I guess.

Yeah but KRK exists so hol up.

MS will need to drop the knees and ask Nvidia and AMD for mercy if they want to Arm Gaming to take off

They need to ask gabeN for that.

 

[GTracing]

Okay, once we are done with the X Elite biz, we can move on to the next great upcoming Snapdragon product.

Snapdragon 8 Gen 4

TSMC N3E node
Oryon CPU
(2 × 4.2 GHz) + (6 × 3.0 GHz)
Adreno 830 GPU
(1.15 GHz)

The custom Oryon CPU architecture, new Adreno 800 series GPU architecture and the 3nm node, Snapdragon 8 Gen 4 is shaping up to be revolutionary...

Or

It will be the Snapdragon 810 all over again!

History doesn't repeat itself, but it often rhymes!

By my estimation, the Oryon cores will beat the x925 in performance and should match or beat it in efficiency, but if Qualcomm doesn't have a medium or little cores, the battery life will be worse. The fact that Qualcomm's last gen chips beat the X Elite in battery life doesn't bode well.

 

[FlameTail]

MS will need to drop the knees and ask Nvidia and AMD for mercy if they want to Arm Gaming to take off

What good will that do?

Nvidia and AMD can supply dGPUs. But I don't think Microsoft intends ARM laptops to be chunky gaming laptops.

ARM laptops will be Thin'n'lights, so as long as they can do casual gaming, it's gonna be fine. To get it to that level, MS will have to throw money ST game developers to port their games, and also work on their own Prism emulation layer.

 

[adroc_thurston]

But I don't think Microsoft intends ARM laptops to be chunky gaming laptops

Hamoa has a very real x8 PCIe g4 root for GPUs and pulls luggable watts anyway so you can.
Well, when aa64 drivers from either dGFX vendor are out.

 

[DavidC1]

Qualcomm CEO Cristiano Amon sounds pumped up just talking about it. “People are going to be incredibly surprised,”

Well, they didn't lie there...

 

[ikjadoon]

Another one:

As reported a few times now, by default ASUS adds a significant automatic underclock when on battery (~1 GHz). Windows Central did not publish any data, but warned users not to use that "Best Efficiency" mode.

Well, I can see why they were reticent to provide data (even though by rights, they should, as that is the default experience): it is a doozy!

Geekbench 6.3 on the ASUS VivoBook S 15
Qualcomm 78 | On AC / plugged in: ~2.4K 1T points / 14.3K nT points
Qualcomm 78 | DC / battery: ~1.0K 1T points (-58%) / 8.5K nT points (-41%)

On battery, the QC 78 in 1T tasks performs like a Broadwell-era Intel Core i7-5650U. That is egregiously slow. Why would ASUS throttle it so much? Some throttling, sure, it is meant to be the "best efficiency". But is ASUS Intel / AMD laptops' on-battery throttling this severe?

I expected 1 GHz to be slow, but seeing the numbers here is wild. A common complaint of today's laptops is the large discrepancy between unplugged vs plugged performance in everyday tasks.

//

I expect, and hope, other OEMs didn't follow ASUS' folly here. Surely the default should be sane throttling (and ideally zero throttling).

 

[adroc_thurston]

But is ASUS Intel / AMD laptops' on-battery throttling this severe

ASUS silent modes do indeed chop a good 1GHz off AMD parts too at least, just that it's a lower impact since you're going from 5GHz to something lower.

 

[MangoX]

ASUS silent modes do indeed chop a good 1GHz off AMD parts too at least, just that it's a lower impact since you're going from 5GHz to something lower.

Correct me if I'm wrong, but concerning that ARM lawsuit. IF ARM wins, does that mean QC will no longer be able to use Nuvia Oryon cores? That QC will have to stick to X cores or build up new cores from the ground up, much like they used to do?

 

[gdansk]

Correct me if I'm wrong, but concerning that ARM lawsuit. IF ARM wins, does that mean QC will no longer be able to use Nuvia Oryon cores? That QC will have to stick to X cores or build up new cores from the ground up, much like they used to do?

They don't want to win which prevents sale they want a settlement which forces Qualcomm to pay more.

 

[ikjadoon]

NBC (Germany) has done some power tests, with somewhat unclear methodology, but it has some interesting data:



However, it would've been more interesting with more laptops. Some of these are as efficient in web browsing, but maybe they fall apart during 4K playback?

ASUS Vivobook S 15: ~11.2 min / WHr
RedmiBook 14 Pro: ~9.9 min / WHr
Dell XPS 13: ~17.4 min / WHr

 

[MangoX]

NBC (Germany) has done some power tests, with somewhat unclear methodology, but it has some interesting data:

View attachment 101489

However, it would've been more interesting with more laptops. Some of these are as efficient in web browsing, but maybe they fall apart during 4K playback?

ASUS Vivobook S 15: ~11.2 min / WHr
RedmiBook 14 Pro: ~9.9 min / WHr
Dell XPS 13: ~17.4 min / WHr

I can't read German so I didnt read the article, just quickly glanced theough it. But those numbers don't make sense at all. Are the AMD/Intel parts using CPU decoding? What codec is being used? VP9/AV1? CPU or GPU decoding? Browser used? Chrome/FF/Edge/Safari? The non x86 numbers align and are where they should be; really hardware decoding of 4k video should really only use around ~1WHr of power with the latest processors, however those x86 numbers don't look right at all. Makes it seems like the CPU is doing all the decoding at that point. Anyone who can read the article can offer more insight?

 

[poke01]

I can't read German so I didnt read the article, just quickly glanced theough it. But those numbers don't make sense at all. Are the AMD/Intel parts using CPU decoding? What codec is being used? VP9/AV1? CPU or GPU decoding? Browser used? Chrome/FF/Edge/Safari? The non x86 numbers align and are where they should be; really hardware decoding of 4k video should really only use around ~1WHr of power with the latest processors, however those x86 numbers don't look right at all. Makes it seems like the CPU is doing all the decoding at that point. Anyone who can read the article can offer more insight?

I’ll test with my i7 9th gen H sku. 15 watts seem very high for 4K playback on YouTube for cpu.

 

[poke01]

The fact that Qualcomm launched their 84 SKU with only 16GB of RAM should tell you that they don’t anyone buying that sku as its low yields.

It also shows that 84 sku is for looking good in benchmarks. This is another joke of launch that’s only better because of Nuvia.

 

[igor_kavinski]

Intel embarrasses themselves once more with their beta silicon.

 

[poke01]

I've tested with this video. On a 2019 Intel 16" i7 MBP. What NBC got is weird.

I get a 12watts total core on a 14nm i7. Those x86 figures should be much lower.

 

[podspi]

I think fixing issues with Numpy and writing some documentation is not Qualcomm's priority at the moment. They have many other important problems.

Usually, people doing some AI development professionally have to use nVidia anyway.

A lot more stuff works in WSL than native windows. It shouldn't be too far off.

 

[FlameTail]

A lot more stuff works in WSL than native windows. It shouldn't be too far off.

How's your Elitebook so far?

 

[SpudLobby]

The question is what was the nature of the issues. If they're software related, then maybe some new GPU driver will come out and give a big boost at some point. If they're hardware related, and they thought "the next stepping or two will fix it" and they were wrong, then that's another matter entirely.

GPU: drivers (and it probably should be bigger for being released now tbh)

NBC (Germany) has done some power tests, with somewhat unclear methodology, but it has some interesting data:
View attachment 101489

However, it would've been more interesting with more laptops. Some of these are as efficient in web browsing, but maybe they fall apart during 4K playback?

ASUS Vivobook S 15: ~11.2 min / WHr
RedmiBook 14 Pro: ~9.9 min / WHr
Dell XPS 13: ~17.4 min / WHr

Interesting.

 

[SpudLobby]

 

[poke01]

View attachment 101506
View attachment 101507

View attachment 101508
View attachment 101509

All this showed me is that Apple is doomed. /s

 

[poke01]

Qualcomm Snapdragon X Elite in der Analyse - Effizienter als AMD & Intel, Apple bleibt aber klar vorne

Notebookcheck-Analyse des neuen Qualcomm Snapdragon X Elite ARM-Prozessor im Asus Vivobook S 15 OLED.

www.notebookcheck.com

What’s interesting to me is that in NBC tested the M4 in Geekbench 6.2 and it scored around 3700 points and 6.2 doesn’t have SME support.

M4 is faster than M3, around 22% in GB6.2.

Really looking forward to Lunar Lake now and how it compares.

Edit: fixed some typos and silliness

 

[Nothingness]

M3 is exactly 18% faster than M4 in GB6.2, matching Geekerwan’s total SPEC improvement perfectly. ~10% clocks and ~8% IPC.

Small typo here

EDIT: And also I find 22% faster, not 18%. 3715/3054

 

[FlameTail]

So do you guys think X2 Elite (Oryon V2) will come in 2025Q3?

 

[Doug S]

They don't want to win which prevents sale they want a settlement which forces Qualcomm to pay more.

And to keep their customers in line by proving that they are willing to go to court to enforce their contracts. Even if Qualcomm ultimately prevails the fact ARM was willing to take it this far would give pause to customers with fewer resources thinking about fighting ARM. Though I really believe ARM thinks they have the upper hand, legally. We'll have to see what happens in court, or if one of them folds at the last minute and comes to a settlement (that often happens if one party believes their case may not be strong enough and they don't want to risk the "unconditional surrender" you have to accept after losing in court)

 

[adroc_thurston]

So do you guys think X2 Elite (Oryon V2) will come in 2025Q3?

It's called Glymur and it's on shelves somewhere H1'26 which is what the roadmap shows anyway.

would give pause to customers with fewer resources thinking about fighting ARM.

but no one at large is fighting ARM at all.
It's just Qualcomm.