An Important Note On The Alder Lake Mobile Power/Performance With Linux 5.19

In the end across dozens of benchmarks there was a clear trend of Linux 5.19 on the Core i7 1280P performing slower than Linux 5.18 stable... But with better power efficiency.

The geo mean across a wide variety of tests both regressed and not came down to roughly a 5% hit overall. All of this benchmarking data in full can be found via this result page.

Looking at the CPU power consumption across all of the benchmarks carried out, there was a clear difference. On average with Linux 5.19 the i7-1280P had a SoC power consumption of 29 Watts compared to 35 Watts with Linux 5.18. At the top-end was also a noticeable difference with Linux 5.18 hitting 107 Watts while on Linux 5.19 Git it topped out at just 77 Watts. The Core i7 1280P is rated for a 28 Watt base power consumption and 64 Watt maximum turbo power consumption.

The i7-1280P CPU temperature over the span of all the benchmarks carried out didn't show any thermal efficiency improvement.

So what's going on with Linux 5.19 for mobile Alder Lake? As bisecting the kernel on this 14 thread / 20 core laptop would still be painfully long process and I am consistently short on time/resources, it was on to thinking about the Linux 5.19 power management changes... Just two months ago for the Linux 5.19 merge window I had written about Linux 5.19 Heavy On Intel Power Management & Thermal Improvements.

One of those changes immediately came to mind and that is the Intel Idle driver support for Alder Lake. Linux 5.19 adds Alder Lake support to the "intel_idle" driver. As that change explained, with Alder Lake and Sapphire Rapids server processors the C1 and C1E power saving states are mutually exclusive with only C1 or C1E power states being exposed at a time, but not both, unlike earlier Intel processors. The Linux kernel prefers C1 for Sapphire Rapids while the Intel engineers set Alder Lake to prefer C1E for power-savings. From initial testing that appears to be at least a contributing role to this power/performance change coming with Linux 5.19. But rather surprising it makes such an impact especially with the magnitude difference in the peak CPU power consumption.

That change did fortunately add a new module parameter of "intel_idle.prefer_cstates=2" if the user wants to choose the C1 states over the default C1E. I am now verifying with that option and just mentioning these results now for any other Phoronix readers that may have Alder Lake laptop(s) and are interested in the power/performance. Unfortunately this is my only Alder Lake P laptop for testing and was a model that I had to buy retail due to lack of vendor interest/support in Linux laptop reviews/testing. If you appreciate these sorts of Linux hardware benchmarks and testing, please consider joining Phoronix Premium or at least disabling your ad-blocker. PayPal tips also accepted.

For those wondering about Alder Lake desktop performance on Linux 5.19, with my testing thus far I haven't yet seen any measurable performance/power differences in the few Alder Lake desktop systems around. Stay tuned there though for more Linux 5.19 desktop benchmarks ahead.