Ice Lake Xeons Will Ramp Up Frequencies Slower, So Linux Is Preparing A Workaround

While being very eager to learn more about Intel next-gen Ice Lake Xeon processors as their move in the server space finally from 14nm to 10nm+, we continue to learn new tid-bits from the open-source Linux kernel activity.

Though the process advancements of Ice Lake allow for power efficiency improvements, the latest kernel activity is pointing to Ice Lake Xeon CPUs actually yielding slower behavior when it comes to ramping up clock frequencies from sleep.

There is a new patch under testing for the Linux power management code from Intel that acknowledges: "On ICX platform, the CPU frequency will slowly ramp up when woken up from C-states deeper than/equals to C1E. Although this feature does save energy in many cases this might also cause unexpected result. For example, workload might get unstable performance due to the uncertainty of CPU frequency."

This issue was reported by Intel's own "kernel test robot" so did appear to come up as an issue in their own internal performance benchmarking, likely in the higher variance in the results if the clock frequencies at times are slow to ramping up or for a lightly-loaded server not as quickly moving to its higher performance states.

Due to this behavior of Ice Lake Xeon server CPUs ramping up slower from C1E (Enhanced Halt State) or lower, the patch allows disabling C1E auto-promotion for these CPUs and disabling of the C1E/C6 sleep states via sysfs. In turn these lower-power sleep states won't be reached, but the CPU can more quickly ramp up in responding to changing (increasing) server load.

The patch is currently in the "bleeding edge" power management tree but still has time for potentially making it into the Linux 5.9 cycle opening next month for those wanting to ensure optimal Ice Lake Xeon performance at the cost of potentially increased power use.