News Snapdragon X Elite pushed past 100W shows us what the CPU can offer on the desktop — almost 4X more power for 10% to 30% more performance

[Admin]

Benchmarks performed on the Snapdragon X Elite dev kit show that having a lot more power doesn't always mean having a lot more speed.

Snapdragon X Elite pushed past 100W shows us what the CPU can offer on the desktop — almost 4X more power for 10% to 30% more performance : Read more

 

[aaronage]

“Nearly quadrupling the power for performance gains in the 10% to 30% region isn’t particularly efficient. However, Geerling did note that the dev kit’s Snapdragon X Elite scored about on par with Apple’s M3 Pro for only “a bit more power.””

Geerling is wrong. It’s not just “a bit more power” lol, 80-100W is more than double.
M3 Pro (6P+6E) uses <30W running a full CPU load (full system power draw from the battery).

 

[Notton]

Another limitation for desktop use is the PCIe lane config
8+4 lanes of 4.0
2+2 lanes of 3.0
It's okay for a mobile chip, but then again, an older R7 7840U has 8+4+4+1+1+1+1=20 lanes of 4.0

 

[bit_user]

the dev kit was only about 10% faster in Geekbench 6 and 28% faster in Cinebench 2024’s multi-threaded test.

Geekbench 6's multithreaded benchmark behaves is so weirdly that I think journalists should stop citing it. We don't know enough what it measures or how, but its multi-core scaling is simply atrocious!

I wonder if the dev kit has customizable power limits. If I had one, I think I'd probably end up wanting to run it somewhere around 45 W.

I wish we had perf/W curves for it, which we could compare to different laptop CPUs. It's interesting how this slide Intel published features a perf/W curve for Lunar Lake, but only selected point measurements for all of the other CPUs.

Source: https://www.tomshardware.com/pc-com...-battery-life-worlds-fastest-mobile-cpu-cores

 

[tennis2]

soooooo, pretty much behaves like any other OC'd chip....

 

[Pro74]

It's well known that performance does not scale linearly with wattage. Obviously Qualcomm is doing something wrong else it would achieve 30-50 percent uplift in single core. But it is not alone in its failure . M3s also fail to see a significant uplift in their single core performance despite the significant increase in tdp from m3 to m3 max.

That performance does not scale linearly with power,has some consequences. The illusion that because , for example a chip that achieves 10 gigaflops per watt at 20 watts is more power efficient than a chip that achieves 13 to 15 gigaflops per watt at 100 watts. X5 increase in power, like in this example was never meant to output x5 in performance .

Hence for example saying that Apple is more efficient at 20 watts than a competitor at 100 watts is tricky and in most cases a flawed impression.

 

[bit_user]

Hence for example saying that Apple is more efficient at 20 watts than a competitor at 100 watts is tricky and in most cases a flawed impression.

That's why we need perf/W curves for the CPUs concerned.

Here's a fun plot I made from composite multithreaded performance data on Raptor Lake and Zen 4.

​

I wish more reviewers got on board with this sort of perf/W testing, because then we could make better tradeoffs on where to set power limits. Obviously, it's workload specific. Having data for the 22 sub-scores in SPEC2017 would cover most bases, at least for non-gaming usage.

 

[usertests]

I'd buy that for a dollar!

 

[defunctup]

“Nearly quadrupling the power for performance gains in the 10% to 30% region isn’t particularly efficient. However, Geerling did note that the dev kit’s Snapdragon X Elite scored about on par with Apple’s M3 Pro for only “a bit more power.””

Geerling is wrong. It’s not just “a bit more power” lol, 80-100W is more than double.
M3 Pro (6P+6E) uses <30W running a full CPU load (full system power draw from the battery).

so this isn't exactly true and not the fairest comparison. the M3 Pro hits a maximum wattage of 73.6w under load according to Notebookcheck. And because the SD X Elite dev kit is a desktop device, noting only the power draw of the M3P on battery is a bit disingenuous. it's well known laptops will scale wattage down on battery compared to plugged it, and yes even modern Macbooks do it. that's why sites like NBC check the full wattage when run under maximum load while on power, which is a much more apt comparison to the X Elite dev kit.

 

[TheHerald]

Hence for example saying that Apple is more efficient at 20 watts than a competitor at 100 watts is tricky and in most cases a flawed impression.

Yeap, that's how we ended up at Intel not being efficient 😂

 

[TJ Hooker]

so this isn't exactly true and not the fairest comparison. the M3 Pro hits a maximum wattage of 73.6w under load according to Notebookcheck. And because the SD X Elite dev kit is a desktop device, noting only the power draw of the M3P on battery is a bit disingenuous. it's well known laptops will scale wattage down on battery compared to plugged it, and yes even modern Macbooks do it. that's why sites like NBC check the full wattage when run under maximum load while on power, which is a much more apt comparison to the X Elite dev kit.

That's a combined CPU + GPU stress test though, and is total system power (including screen at max brightness). This article and the person you are responding to are talking about CPU-only benchmarks/power.

If you look at the CPU-only results in your link (that also use an external monitor), they measure roughly 30-40W.

 

[aaronage]

so this isn't exactly true and not the fairest comparison. the M3 Pro hits a maximum wattage of 73.6w under load according to Notebookcheck. And because the SD X Elite dev kit is a desktop device, noting only the power draw of the M3P on battery is a bit disingenuous. it's well known laptops will scale wattage down on battery compared to plugged it, and yes even modern Macbooks do it. that's why sites like NBC check the full wattage when run under maximum load while on power, which is a much more apt comparison to the X Elite dev kit.

I have an M3 Pro machine. If I run a full CPU load (Cinebench, multi-threaded compile etc.) the total system power draw from battery (reported by AlDente) is <30W. It can use double that when both the CPU and GPU are maxed out.
I don’t know how notebookcheck are measuring it - maybe they report from the wall including losses (chargers are only 80-90% efficient IIRC)?

 

[bit_user]

I don’t know how notebookcheck are measuring it - maybe they report from the wall including losses (chargers are only 80-90% efficient IIRC)?

I'm not sure which of their chargers are GaN-based, but here's a review of one that shows efficiency above 90% for everything about 5V.

Review of New Apple 70W GaN USB-C Charger (A2743)

Several days ago, we posted the teardown video of the brand new Apple 70W GaN charger, you can click the related articles to check it. And as promised, we gonna test the charging performance of this charger today.

www.chargerlab.com

You can get a rough clue about charger efficiency by its size vs. power ratio. The only way to make a powerful charger really small is by making it very efficient. Otherwise, it'd get too hot.