Discussion Qualcomm Snapdragon Thread

[SpudLobby]

Look also at how low Qualcomm can go for platform power in ST — at a reduced but still functional performance of about 1/3 of each peak ST, Qualcomm can drop down to 2W whereas AMD/Intel are at around 5. Telling about the fabrics of each and the scalability of the core.

QC may not have real E cores this round but I bet the battery life will still be great.

 

[gdansk]

QC may not have real E cores this round but I bet the battery life will still be great.

Yep, they don't need an e core to fight MTL and Zen 4.
They need it to fight Apple, indirectly, in phones.

 

[FlameTail]

View attachment 96340

66% faster than 8945HS in PCMARK10. If some sort of benchmark manipulation trickery isn't going on, that's quite impressive.

I think Adroc mentioned that the Zephyrus they are using for the comparisons is running single channel (64 bit) RAM....

 

[FlameTail]

Yep, they don't need an e core to fight MTL and Zen 4.
They need it to fight Apple, indirectly, in phones.

No, they'll need it next generation even to fight AMD/Intel.

AMD is set to make a colossal leap in their P-core power efficiency with Zen 5. And Intel is finally getting rid of their node disadvantage, and pursuing E-cores.

 

[FlameTail]

Look also at how low Qualcomm can go for platform power in ST — at a reduced but still functional performance of about 1/3 of each peak ST, Qualcomm can drop down to 2W whereas AMD/Intel are at around 5. Telling about the fabrics of each and the scalability of the core.

QC may not have real E cores this round but I bet the battery life will still be great.

As good as Zen 5 (Strix Point) might be, it won't beat X Elite in this aspect. Even Adroc has admitted that X Elite will have fabric/idle power advantage.

 

[gdansk]

AMD is set to make a colossal leap in their P-core power efficiency with Zen 5

So some say, but I think even the normal improvements and going from 8->12 cores is enough to make it a threat to SDXE.

 

[FlameTail]

Look also at how low Qualcomm can go for platform power in ST — at a reduced but still functional performance of about 1/3 of each peak ST, Qualcomm can drop down to 2W whereas AMD/Intel are at around 5. Telling about the fabrics of each and the scalability of the core.

QC may not have real E cores this round but I bet the battery life will still be great.

Skip to 13:00

Gerard Williams talks about the X Elite having only performance cores, and no efficiency cores.

 

[FlameTail]

Redditor explains power measurements:

https://www.reddit.com/r/hardware/comments/1bui8z9/comment/kxy12s3

If M3 is consuming as much as 10W for ST, then 13W for X Elite ST isn't too bad.

 

[FlameTail]

https://twitter.com/x/status/1775751563891650831

Olrak wants an Android tablet with X Elite

 

[MarkizSchnitzel]

Doesn't matter how good (performant/efficient) Intel and AMD's next gen offerings are, WoA will continue to exist and grow.

I believe that is the only things that matters.
I just want a fanless but speedy browsing machine with nice screen and keyboard and good battery life. And all that in midrange.

 

[FlameTail]

Qualcomm is flying tech influencers to their HQ to show off the Snapdragon X Elite:

 

[FlameTail]

I am kinda disappointed with the graphics of X Elite.

M2/M3 level of performance at M2 Pro/M3 Pro levels of power consumption.

I want them to beat the base M chip's graphics performance, while having similar efficiency.

We know Qualcomm can do it. Their Adreno is the leader in mobile GPUs.

The reason the X Elite lags behind M2/M3's GPU efficiency is probably because they did not allocate sufficient transistor budget to it. It explains how they achieved the miniscule ~170 mm² die size.

X Elite's GPU efficiency right now is good enough to beat Meteor Lake and Hawk Point, but I suspect next gen Lunar Lake and Stric Point might probably beat it. But it's certainly nowhere at Apple's level.

The impressive thing about Apple Silicon is every part of it is efficient. Not just the CPU (which is most talked about), but also the GPU, NPU, Media Engine, display engines, thunderbolt ports etc.... Apple is willing to splurge with regards to the transistor budget. A single Apple Thunderbolt controller in an M chip takes up as much area as a CPU P-core!

I wish future Snapdragon X processors would actually splurge on the GPU transistor budget, so we can do some actual power efficient gaming on it, and it would make it fit to put in tablets.

However, looking at the rumour mill, it sounds rather grim. Revegnus (in one of his last leaks), said X Elite Gen 2 will use an overclocked Adreno 830 GPU.

FWIW, Adreno 830 is the GPU powering the upcoming Snapdragon 8 Gen 4. It is 10% faster than M2's GPU, and the same performance as the GPU in the X Elite 80W TDP.

 

[NTMBK]

SAMSUNG ELECTRONICS CO., LTD. Galaxy Book4 Edge vs ASUSTeK COMPUTER INC. ROG Zephyrus G14 GA403UI_GA403UI - Geekbench

browser.geekbench.com

Bit better but it's gonna be a lot more expensive than the Asus laptop. People really wanna pay the premium for better browsing speed?

A lot of people do the majority of their day to day job inside a web browser these days. "Better browsing speed" means they can do their job faster and more easily.

 

[adroc_thurston]

I am kinda disappointed with the graphics of X Elite.

don't worry you'll feel worse soon

 

[gdansk]

And ARM added instruction(s) which match JavaScript behavior to enable faster JITs: https://developer.arm.com/documentation/dui0801/h/A64-Floating-point-Instructions/FJCVTZS

With all its complexity is there even a single x64 instruction to do that? cvttsd2siq is close but still compare the implementations of TruncateDoubleToI in the arm64 and x64 macroassemblers in V8. A feature check and a single instruction vs all that crap. They just aren't innovating at the same rate.

 

[NTMBK]

And ARM added instruction(s) which match JavaScript behavior to enable faster JITs: https://developer.arm.com/documentation/dui0801/h/A64-Floating-point-Instructions/FJCVTZS

With all its complexity is there even a single x64 instruction to do that? cvttsd2siq is close but still compare the implementations of TruncateDoubleToI in the arm64 and x64 macroassemblers in V8. A feature check and a single instruction vs all that crap. They just aren't innovating at the same rate.

I thought JavaScript has this behaviour because that's how it works by default on x86? Is this not just replicating some old SSE truncate behaviour?

 

[gdansk]

I thought JavaScript has this behaviour because that's how it works by default on x86? Is this not just replicating some old SSE truncate behaviour?

It's close to cvttsd2siq but x64 version ends up being more instructions. At least 4 including an often untaken branch.

Although it was much worse on ARM before that which is probably why they added it.

Also to be clear I have no idea about assembly I just want faster JavaScript engines.

 

[Nothingness]

It's close to cvttsd2siq but x64 version ends up being more instructions. At least 4 including an often untaken branch.

Although it was much worse on ARM before that which is probably why they added it.

Gory details:

v8/src/compiler/backend/x64/code-generator-x64.cc at main · v8/v8

The official mirror of the V8 Git repository. Contribute to v8/v8 development by creating an account on GitHub.

github.com

v8/src/codegen/arm64/macro-assembler-arm64.cc at main · v8/v8

The official mirror of the V8 Git repository. Contribute to v8/v8 development by creating an account on GitHub.

github.com

 

[FlameTail]

don't worry you'll feel worse soon

 

[FlameTail]

And ARM added instruction(s) which match JavaScript behavior to enable faster JITs: https://developer.arm.com/documentation/dui0801/h/A64-Floating-point-Instructions/FJCVTZS

With all its complexity is there even a single x64 instruction to do that? cvttsd2siq is close but still compare the implementations of TruncateDoubleToI in the arm64 and x64 macroassemblers in V8. A feature check and a single instruction vs all that crap. They just aren't innovating at the same rate.

Ooh. Very interesting.

 

[igor_kavinski]

Ooh. Very interesting.

No. Not really. I smell Google. They stand to gain the most from performant Javascript. They want you to never leave ChromeOS or Android or Gmail or Google Docs. You wanna know how often I have to kill a Win32 process due to errant behavior consuming too much CPU? Almost never. How often do I kill a tab in a browser due to errant JS script eating CPU cycles? Almost every day.

 

[Nothingness]

No. Not really. I smell Google. They stand to gain the most from performant Javascript. They want you to never leave ChromeOS or Android or Gmail or Google Docs. You wanna know how often I have to kill a Win32 process due to errant behavior consuming too much CPU? Almost never. How often do I kill a tab in a browser due to errant JS script eating CPU cycles? Almost every day.

This also happens with Firefox 😉
Browsers and their JS engines have become part of the most complex software stacks.

 

[FlameTail]

Electron

 

[eek2121]

Yeah but that doesn't explain the low 2400 ST score. Qualcomm has shown that X Elite can do 2700 in a 23W device TDP envelope.

So this Lenovo device is either a <10W fanless device or a fanned laptop but still in testing/debugging phase.

We really need independent reviews.

Btw why doesn't this thread have the functionality of upvoting/downvoting comments?

So it was you doing that. Those arrows are not for what you think they are for FYI.

https://twitter.com/x/status/1775751563891650831

Olrak wants an Android tablet with X Elite

To be fair, the one thing that could pry me away from my iPad would be a 7” tablet with the x elite running Android so I agree with him.

 

[Ghostsonplanets]

Isn't the 8G4 basically a Phoenix for Phones with the 2P + 6E Phoenix Cores. While also having the latest updated IPs from QCOM, specially the shiny new Adreno 8 series uArch?

I don't see much reason for X Elite on the Android space due to this.