4

January 15, 2018 update

I created a 90 second Conky .gif showing "normal" behavior during start up:

  • A few seconds after boot I start Conky.
  • A few seconds later I start Peek which begins recording .gif after 3 seconds when system up-time is about about 30 seconds.
  • When .gif starts I load Firefox 10 tabs and two windows, one window with video (flash player)
  • For first minute up up-time, .gif shows frequency high in turbo mode and CPU load moderate.
  • cronjobs, for example daily emailed backup, run at @reboot time making single core spike to 100%.
  • After a minute frequencies reduce to non-turbo boost below 2.6 GHz which I deem to be "normal".
  • .gif ends when system up-time is 2 minutes and 10 seconds.

Normal start-up .gif:

Conky startup.gif

Although this is under the same kernel 4.14.13 it doesn't explain the abnormal behavior on January 13, 2018 when frequencies stayed in turbo-boost mode during the entire session.

January 14, 2018 update

I rebooted with Kernel 4.14.13 today and it is behaving like 4.14.12 yesterday. I can't explain the difference and have no objections with this question be close voted.

High CPU % with Low Frequency vs. Low CPU % with High Frequency

Both these results are using FireFox with 10 tabs open in Non-Full Screen on laptop built-in 1080p display and FireFox with one tab open in Full Screen streaming a movie on External TV connected via ThunderBolt 3 USB-C to HDMI adapter. All tests were on a Skylake i7 6700HQ.

Kernel 4.14.12

All kernels sampled 4.4, 4.10 and 4.14 (up until version 4.14.13) exhibited High CPU % with Low Frequency.

CPU High Frequency Low.gif

Kernel 4.14.13

Whilst trying different kernels to watch Meltdown security hole impact I noticed an unrelated change where clock frequency was running at Turbo speeds constantly with a 10 degree increase in temperatures.

CPU Low Frequency High.gif

What Changed

The only difference is booting Kernel version 4.14.12 (High CPU, Low frequency) and kernel version 4.14.13 (Low CPU with High Frequency).

Ubuntu 16.04.3 LTS is used in both tests with no apt-get changes in-between.

In both tests the nVidia GTX 970M is disabled and only the Intel HD530 integrated graphics are used. When nVidia was activated on Kernel 4.14.4 the CPU % didn't seem to decrease all that much but moving Windows on desktop wasn't nearly as smooth as the Intel iGPU.

My other laptop has an Ivy Bridge 3630QM CPU and exhibits very different behavior in the three years I was using it. CPU % was slightly higher because it was a weaker processor but frequency would bounce between low and high all the time with most time spent in the middle. I think the frequency behavior was much preferable but, that is subjective. I'll create some Conky .gif's later for the 3630QM CPU if someone is interested.

Questions

Is it better to have Low CPU with High Frequency or High CPU and Low Frequency?

Battery use isn't really a concern as this 17" laptop doesn't go outside since it doesn't fit into the coat pocket as easily as the 5.5" 1080p smartphone. It would still be good to know if Low CPU % or Low Frequency is better for battery life?

closed as too broad by Thomas Dickey, Fox, maulinglawns, G-Man, Archemar Jan 15 '18 at 10:57

Please edit the question to limit it to a specific problem with enough detail to identify an adequate answer. Avoid asking multiple distinct questions at once. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.

1

The answer to this question will really depend on use case, but remember post Haswell The core speeds can ramp up based on need per core, and that the "Uncore" or non-CPU bound (memory bound as an example) can ramp up the frequency without impacting the power budget of the core.

In general, if the latency in ramping up does not effect your interactive feel you want the frequencies as low as possible.

The turbo speed of CPUs is limited by the power and thermal budgets of the package, and if most of your cores are running at a low frequency a single thread can turbo higher and faster for longer.

Also note that some functions like AVX2 require significantly more power than integer operations and on most chips will reduce the CPU frequency below the base frequency if all cores are at a high speed.

I am not familiar with the tool you are using, but due to this change some are looking at the TSC frequency and not the core frequency so this may not be observed.

Unfortunately due to the dynamic nature of Turbo Boost and some unseen PID functions related to core and package temperature the %used number aren't very useful.

Targeting acceptable latency (which is much better in modern iterations of the scaling drivers in linux) and lowest temperature will most likely lead to improved single threaded performance.

I would check

$ cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_governors

And:

/sys/devices/system/cpu/*/cpufreq/scaling_governor

To make sure you aren't on the performance governor which will sacrifice individual thread performance for latency.

  • On my i7 6700HQ the governor available scaling options are powersave and performance. All 8 CPUs are set to powersave which is the default. My older Ivy Bridge laptop was setup the same way and it's the default in Ubuntu. Conky is displaying the frequency of the highest operating core as when a single thread spikes up to 100% you'll see the solo frequency displayed spike up as well. – WinEunuuchs2Unix Jan 14 '18 at 15:39

Not the answer you're looking for? Browse other questions tagged or ask your own question.