| 1 |
On Thursday, 5 March 2020 00:28:21 GMT Walter Dnes wrote: |
| 2 |
> I've cobbled together a script to select cpu governors and speeds. |
| 3 |
> One weird thing I've noticed is that reported cpu speed doesn't quite |
| 4 |
> match the selected speed. E.g. on my machine (yours will vary)... |
| 5 |
> |
| 6 |
> cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies |
| 7 |
> |
| 8 |
> ...shows avalable speeds... |
| 9 |
> |
| 10 |
> 3001000 3000000 2900000 2800000 2700000 2600000 2500000 2400000 2300000 |
| 11 |
> 2200000 2100000 2000000 1900000 1800000 1700000 1600000 |
| 12 |
> |
| 13 |
> ***IMPORTANT*** "userspace" governor must be present. |
| 14 |
> |
| 15 |
> /sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq reports speeds |
| 16 |
> very close to, but slightly lower than the selected values, and they also |
| 17 |
> seem to jump around a bit. What gives? |
| 18 |
|
| 19 |
My reported CPU frequency is also slightly lower than the values reported in |
| 20 |
'scaling_available_frequencies' and it fluctuates as the load on the CPU |
| 21 |
varies. I take these values to be rounded upper limits, rather than what the |
| 22 |
CPU will be pegged at. |
| 23 |
|
| 24 |
MoBo firmware plus kernel options seem to control (limit) the CPU frequency. |
| 25 |
I was installing Gentoo on a laptop, using a minimal CD and the CPU thermal |
| 26 |
cut out would kick in every time I tried to install packages shutting down the |
| 27 |
PC. The frequency at the time reached 3600-3900MHz. After I completed the |
| 28 |
install by limiting jobs and using an external cooling fan, I booted with my |
| 29 |
own newly configured kernel. I felt disappointed to notice the frequency |
| 30 |
would never go above 2500, no matter what scaling_governor I used and what max |
| 31 |
frequency I selected. I also tried using sys-power/cpupower to manipulate |
| 32 |
governors and frequencies, but nothing would change this hard limit of |
| 33 |
2500MHz. |
| 34 |
|
| 35 |
I can't recall what the LiveCD was missing in the kernel/packages - it might |
| 36 |
have been acpi-freq. This is the current state of affairs: |
| 37 |
|
| 38 |
$ cpupower frequency-info |
| 39 |
analyzing CPU 0: |
| 40 |
driver: acpi-cpufreq |
| 41 |
CPUs which run at the same hardware frequency: 0 |
| 42 |
CPUs which need to have their frequency coordinated by software: 0 |
| 43 |
maximum transition latency: 4.0 us |
| 44 |
hardware limits: 1.40 GHz - 2.50 GHz |
| 45 |
available frequency steps: 2.50 GHz, 2.10 GHz, 1.80 GHz, 1.40 GHz |
| 46 |
available cpufreq governors: ondemand userspace performance schedutil |
| 47 |
current policy: frequency should be within 1.40 GHz and 2.50 GHz. |
| 48 |
The governor "schedutil" may decide which speed to use |
| 49 |
within this range. |
| 50 |
current CPU frequency: Unable to call hardware |
| 51 |
current CPU frequency: 1.31 GHz (asserted by call to kernel) |
| 52 |
boost state support: |
| 53 |
Supported: yes |
| 54 |
Active: yes |
| 55 |
|
| 56 |
Anyway, since I couldn't push it above 2500MHz, I've left it alone running on |
| 57 |
schedutil and it never overheats or cuts out. |
Archives