Earlier this week I published Linux 3.9 through Linux 4.9 kernel benchmarks looking at the raw performance of various subsystems when testing each of the major kernel releases as far back as this Core i7 Haswell system was supported. From that same system, today is a look at testing the kernels going back to Linux 3.11 when Haswell graphics support was first in good shape for this Core i7 4790K box while looking at the raw power consumption and performance-per-Watt for these 19 major kernel releases.
From Linux 3.11 (the first major release where the HD Graphics 4600 were playing nicely on this Ubuntu 16.10 system) to Linux 4.9 (latest Git) each major kernel release was tested. While running a smaller sub-set of benchmarks then what was used for the big performance comparison earlier this week, the power consumption was monitored by the Phoronix Test Suite along with generating performance-per-Watt metrics. The AC system power consumption was polled by the Phoronix Test Suite using the WattsUp Pro USB power meter; from the PTS side it's just a matter of setting the MONITOR=sys.power and PERFORMANCE_PER_WATT=1 environment variables to activate the support during the benchmarking process for complete automation and reproducibility.
The same system was used throughout testing and was once again the Core i7 4790K, Gigabyte Z97-HD3P, 16GB of RAM, 120GB OCZ TRION 100 SSD, and the integrated HD Graphics 4600. Reiterating from the earlier article, all kernels were obtained from the Ubuntu Mainline Kernel PPA for being close to a vanilla build and no other hardware/software changes were made during the testing process. The reported CPU/hardware differences from the included system table just come down to the kernels exposing different hardware information depending upon the release, etc.
So if you've been curious if the Core i7 desktop power efficiency on Linux has improved at all in the last three years since Linux 3.11, keep reading.