It's honestly quite impressive that they're so close!

That's pretty cool, Intel's come a long way for iGPU's, bring the competition on.

I'm, especially taking note of the rather poor power scaling both exhibit. In some cases adding like 50% more power can double the performance because of the high base power draw, but that's clearly not the case here. Scaling beyond 20W looks quite weak.

Also lol @ 1% lows of MHW.

NostrasI'm, especially taking note of the rather poor power scaling both exhibit. In some cases adding like 50% more power can double the performance because of the high base power draw, but that's clearly not the case here. Scaling beyond 20W looks quite weak.

Also lol @ 1% lows of MHW.

It has never been a linear line. however the Z2E performs better under lower power than the Intel.

would be important to do a real test ( TPU??), in particular with real life power consumption.

I read your test of the AI+. The 37W was consuming the same as the 30W while performance were better.

So the "17W", "30W" are not necessarily accurate, this is just instruction you give and you are not sure how it is actually used.
Also, 17W is the SOC package right? So RAM would be counted in the Intel system, but not in the AMD system i guess?

edit:
in www.techpowerup.com/review/claw-8-ai-a2vm/14.html
the power scaling of the lunar lake was bad at 15-25W, already below the ZOTAC ZONE and the ROG ALLY.

However, it was better at 10W, very close at 30W, >10% better at 37W (without incurring more real life power):
"power usage at 37 W doesn't increase compared to using the 30 W power profile."

Did a bit of investigation, comparing the review of the CLAW 8 AI+ witht he Zotac Zone.

at claimed "30W", the Claw lasted 103min, which is 1.72 hours. With its massive 80Wh battery, it means it consumes 46.5 W/hour (granted, there is a "large" 8" screen etc to power up, not just the SOC).
at claimed "30W", the Zotac Sonz lasted 57min, which is 0.95 hours. With its 48.5 Wh battery, it means it consumes 51W/hour. It has only a 7" screen. That s more than 4.5W/hour discrepency between both., while the screen of the Claw is clearly bigger than the Zotac Zone...

So it would seem that 30W claimed here is not exactly comparable with 30W claimed there.

OF course, comparing on different device is not easy. But this test would be much more relevant on the MSI Claw A8!

bgxedit:
in www.techpowerup.com/review/claw-8-ai-a2vm/14.html
the power scaling of the lunar lake was bad at 15-25W, already below the ZOTAC ZONE and the ROG ALLY.

Interesting review.

bgxDid a bit of investigation, comparing the review of the CLAW 8 AI+ witht he Zotac Zone.

at claimed "30W", the Claw lasted 103min, which is 1.72 hours. With its massive 80Wh battery, it means it consumes 46.5 W/hour (granted, there is a "large" 8" screen etc to power up, not just the SOC).
at claimed "30W", the Zotac Sonz lasted 57min, which is 0.95 hours. With its 48.5 Wh battery, it means it consumes 51W/hour. It has only a 7" screen. That s more than 4.5W/hour discrepency between both., while the screen of the Claw is clearly bigger than the Zotac Zone...

So it would seem that 30W claimed here is not exactly comparable with 30W claimed there.

SoC level might be comparable. But Lunarlake has a platform level advantage. If you compare the power level differences it's consistently little over 4W in all the TDP levels. Meaning Lunarlake gets bigger advantage the lower TDP it's set at, because the 4W difference is larger.

And it's more than just memory, as laptop memory uses 1-1.5W in vast majority of scenarios.

Is the Intel chip able to be undervolted? The AMD Z1E certainly can't, which stinks. Probably the same for the Z2E.

Once these handhelds are built on the 3nm or even 2nm process, they're going to be awesome.
Right now, if you want good framerates, you have to play older games. Loss Scaling does help too, but there's needs to be free GPU overhead.

CyberCTIs the Intel chip able to be undervolted? The AMD Z1E certainly can't, which stinks. Probably the same for the Z2E.

Once these handhelds are built on the 3nm or even 2nm process, they're going to be awesome.
Right now, if you want good framerates, you have to play older games. Loss Scaling does help too, but there's needs to be free GPU overhead.

i wish i know how.

on my 8850, I could /2 the power consumption. But i cant touch anything on lunra lake. Even intel very bad own software does not work at all on lunar lake.

Throttlestop gives you info, but you cant set anything manually sadly, not even the turbo. Which is sad because clearly, the algorithm from intel is bad at setting power.
I did some investigation wih my new laptop, and I could only improve a bit things by setting some hidden microsoft powersetting.

It would Turbo the shit out of P cores first, which is a stupidity in terms of power consumption. I wish I could cap the P cores to someting reasonable, at least when i am not playing game.