AMD Ryzen AI 300 Series Dominates Intel Core Ultra 7 Lunar Lake Performance For Linux Developers & Creators

[phoronix]

Phoronix: AMD Ryzen AI 300 Series Dominates Intel Core Ultra 7 Lunar Lake Performance For Linux Developers & Creators

Earlier this week I delivered initial Intel Xe2 Lunar Lake graphics benchmarks on Linux while today the focus is on Lunar Lake's CPU performance. The Xe2 graphics performance under Linux was disappointingly slow with it performing even worse than Meteor Lake while RDNA3.5 graphics led. Intel has been investigating the Xe2 Linux graphics performance but I haven't heard any updates yet. Today the attention is on the Lunar Lake CPU side under Linux and it too isn't looking too good. The performance of this 8-core Core Ultra 7 256V SoC is poor in real-world multi-threaded scenarios and the performance-per-Watt is only compelling in a subset of workloads. The AMD Ryzen AI 9 365 and AMD Ryzen AI 9 HX 370 Zen 5 SoCs tended to deliver the superior performance and power efficiency under Linux.

https://www.phoronix.com/review/core-ultra-7-lunar-lake-linux

[avis]

54W 12 cores/24 threads dominate 37W 8 cores/8 threads?

I would think so. It looks like the weight categories are quite different, don't you think?

What about battery life though? What about video playback life though?

What about idle power consumption? AFAK it's close to 4W for Ryzen HX 370 and roughly 0.5W for Intel Core 256V.

These are laptop CPUs, not workstations, Michael. And people rarely use them for heavy MT tasks, unless you don't mind a high-pitched noise close to your ears.

[milkylainen]

Feels like Intel neutered this platform in the name off efficiency.
And while way more efficient and an amazing improvement it is still saddening (from a competition perspective) to watch most of the graphs.

Instead of 1/3 of the power to performance efficiency it's "just" 1/2...

[eszlari]

Some of these results look suspicious:

In the audio encoding benchmarks (I guess all single threaded?), how is Zen5 not even able to beat Tiger Lake?

[darkbasic]

Has the ASUS Zenbook S 16 a 28W TDP? What other Ryzen AI alternatives are available on the market with higher TDP?

[David Huang]

Some of these results look suspicious:

In the audio encoding benchmarks (I guess all single threaded?), how is Zen5 not even able to beat Tiger Lake?

The heterogeneous scheduling of Strix Point SoCs doesn't work well under Linux at the moment. I've been daily driving the very same laptop tested in the article for several days using latest 6.11 kernel with all known kernel patches applied, and it still priotizes Zen5c cores for a lot of the single-threaded scenarios.

[David Huang]

54W 12 cores/24 threads dominate 37W 8 cores/8 threads?

I would think so. It looks like the weight categories are quite different, don't you think?

What about battery life though? What about video playback life though?

What about idle power consumption? AFAK it's close to 4W for Ryzen HX 370 and roughly 0.5W for Intel Core 256V.

These are laptop CPUs, not workstations, Michael. And people rarely use them for heavy MT tasks, unless you don't mind a high-pitched noise close to your ears.

The HX370 in Zenbook S16 has a maximum cTDP of 28W only.

[willmore]

This Lunar Lake chip needs to be treated as a quad core. It's only got four 'big' cores and lacks HT. The E cores fill in the gap for the missing HT. Up against 8 core 16 thread chips, it's going to only survive on single threaded performance and low power. That's one win the 4P/4E vs 4P w/HT gets is with the low power options the 4E cores open up. For times when the 4P cores aren't needed, the 4E cores can take over and bright much better power efficiency.

But it boils down to this competing with quad core procesors, not octo core.

[guspitts]

1. Core Ultra 7 256V: 3222 Joules per run
2. Ryzen AI i HX 370: 2598 Joules per run

The Core Ultra 7 256V was consuming less power than the other laptop CPUs which is nice if you are doing a lot of code compilation on battery while traveling or other environments, but the AMD Zen 5 SoCs can also be reduced to a more power-saving mode if desired.

Michael I think looking at Watts rather than Joules leads you to wrong conclusion. Not only was the Intel CPU slower to compile the kernel, but it used more energy than the AMD. And the total amount of energy (i.e., Joules) is what the battery limits you to.

So one would be able to compile more, not fewer, kernels on a single battery charge with the AMD laptop than the Intel (and as a bonus, it will also take you less time).

[chuckula]

Arrow Lake H should compete nicely with the larger Strix parts if that's your intended workload.

I'd take Lunar Lake any day of the week for a practical laptop though.