A Quick Look At Feral GameMode / P-State Powersave vs. Performance

With today's release of the Feral GameMode library/daemon of course I am running some benchmarks... Here is some initial numbers while this has spurred some fresh benchmarks looking at the P-State vs. CPUFreq performance and their respective frequency scaling governor options.

I will be doing some fresh Linux gaming benchmarks using P-State/CPUFreq and their governor options while testing with both Radeon and GeForce graphics cards. First up is the Intel Core i7 8700K tests to be followed by similar AMD Ryzen tests with CPUFreq.

As covered in this morning's article, the GameMode project at this point is mainly about ensuring while games are running on the system, the CPUFreq/P-State governor is set to the "performance" governor rather than "powersave" or "ondemand" depending upon your distribution and CPU configuration. There isn't any "magical" performance tweaks or other work at this point.

These initial one-page tests on Ubuntu were with a Core i7 8700K and NVIDIA GeForce GTX 1080 graphics card while the featured article coming up is with a GTX 1060 as well as an RX 580 and RX Vega 64, followed by AMD Ryzen tests. Intel tests are being done with all the governor options on both CPUFreq and P-State.

But short of digging through the code and looking to see for any other tweaks, running some quick benchmarks confirm that the Feral GameMode performance comes down to being the same as P-State with the performance governor... Compared to Ubuntu, for example, defaulting to P-State powersave with modern Intel CPUs. The benefit of using GameMode though is the hope more performance optimizations will come in the future and that when you aren't gaming, the system will automatically switch back to using the power-savings governor configuration.

With Feral's upcoming Rise of the Tomb Raider port being the first game supporting the GameMode library by default, for now after installing the client library and systemd service you need to manually preload the libgamemodeauto.so library to see any change. For better or worse, GameMode isn't relying upon any program name whitelisting or simply changing mode based upon if Steam is running or anything like checking for CPU and GPU utilization as a more universal heuristic.



But as shown in past CPU frequency scaling comparisons, it largely depends upon the game and CPU as to whether you will find benefits from switching to the performance governor...

With this testing, the CPU temperature and AC system power consumption are also being monitored during the benchmarking process...

Using the "powersave" governor leads to the CPU during game being just about one degree lower for this Core i7 8700K.

While during the benchmarking process, using the performance governor rather than powersave led to an average AC power increase of about seven Watts while the peak power consumption was the same.

Stay tuned to a more complete analysis with fresh benchmarks of CPUFreq/P-State governors for Linux gaming later today or tomorrow on Phoronix.