The Idle Power Use Of The Past 19 Linux Kernel Releases
This morning I published the Power Consumption and Efficiency Of The Linux Kernel For The Last Three Years article containing power consumption data for an Intel Haswell system going back to the Linux 3.11 kernel through Linux 4.9 Git. Those were some interesting power consumption numbers under load while here are the idle numbers.
The idle tests were still running this morning so I opted to post them later since they're interested in their own right. The same i7-4790K system was used for benchmarking all of these kernels from Linux 3.11 to Linux 4.9 (25 October Git). No other changes were made during the testing process. Each kernel was freshly booted to the Unity desktop and then launched the idle power consumption test for a period of three minutes while monitoring the AC power draw as reported by the WattsUp Power meter. Automating this with the Phoronix Test Suite: MONITOR=sys.power phoronix-test-suite benchmark idle.
Here are those idle power results for the Haswell system:
The results fluctuated by a few Watts over the past three years of kernel releases but overall wasn't too dramatic for this Core i7 4780K system. There were spikes though to report from Linux 4.3 through Linux 4.5, but fortunately was addressed. The minimum power draw achieved was actually with Linux 4.6 but sadly didn't carry over through to the newer kernels. There's been a lot of work recently on P-State/CPUFreq, making use of scheduler utilization data, and other power management efforts but at the end of the day for this i7-4790K system it hasn't yielded any real shock.
The idle data in full via this OpenBenchmarking.org result file. See the rest of the kernel power consumption benchmarks from this morning if you haven't already to see how the kernels compare under load and the performance-per-Watt.
The idle tests were still running this morning so I opted to post them later since they're interested in their own right. The same i7-4790K system was used for benchmarking all of these kernels from Linux 3.11 to Linux 4.9 (25 October Git). No other changes were made during the testing process. Each kernel was freshly booted to the Unity desktop and then launched the idle power consumption test for a period of three minutes while monitoring the AC power draw as reported by the WattsUp Power meter. Automating this with the Phoronix Test Suite: MONITOR=sys.power phoronix-test-suite benchmark idle.
Here are those idle power results for the Haswell system:
The results fluctuated by a few Watts over the past three years of kernel releases but overall wasn't too dramatic for this Core i7 4780K system. There were spikes though to report from Linux 4.3 through Linux 4.5, but fortunately was addressed. The minimum power draw achieved was actually with Linux 4.6 but sadly didn't carry over through to the newer kernels. There's been a lot of work recently on P-State/CPUFreq, making use of scheduler utilization data, and other power management efforts but at the end of the day for this i7-4790K system it hasn't yielded any real shock.
The idle data in full via this OpenBenchmarking.org result file. See the rest of the kernel power consumption benchmarks from this morning if you haven't already to see how the kernels compare under load and the performance-per-Watt.
4 Comments