CPU frequency decreases under load, rises when idle

terdon · 03-27-2012, 10:48 AM

Hello all, I have a strange problem. As it says in the title, my cpu frequency will decrease when the system is under load. The problem seems to be that the BIOS reports different max cpu freqs depending on whether the system is under load. The details given below show the "performance" governor but I get similar behaviour with userspace and ondemand. In all cases, I cannot override the BIOS limit if that is set to the low value (see below).

I get the same problem using either 3.0.0-1-amd64 or 3.2.0-2-amd64 kernels. I have just updated my BIOS and saw no difference. I have tried disabling dynamic frequency by disabling Intel SpeedStep Technology in the BIOS. I got a similar situation. In this case, the CPU was reporting a constant frequency of 2.67 GHz but minor tasks (eg firefox, conky, emacs) would occupy >20% of CPU and the CPU usage would quickly reach 99%. I mean that before running "stress" or similar, firefox was sitting happily at ~6% CPU, As soon as I started stressing the system, firefox shot up to ~20% despite the CPU freq apparently not changing. This seems to imply that the CPU freq is misreported and disabling Intel SpeedStep does not stop the decrease in freq, just stops the system from reporting it.

So, either using dynamic frequency or not, my system becomes unusable when even under minor processor loads. Does anyone have any ideas how to fix this? I seem to recall there are known regressions in the 3.0-3.2 kernels that could affect CPU frequency control. Anybody know if that is true? Should I just wait for 3.3?

System Specs:
Dell M4500
Intel Core i7 M 620 @ 2.67GHz
4GB RAM
LMDE, 3.2.0-2-amd64
BIOS: Dell Inc. A10 11/30/2011

Details when using "performace" governor:
1. System idle:
cpufreq-info output (for one processor):

Code:

cpufrequtils 007: cpufreq-info (C) Dominik Brodowski 2004-2009
Report errors and bugs to cpufreq@vger.kernel.org, please.
analyzing CPU 0:
  driver: acpi-cpufreq
  CPUs which run at the same hardware frequency: 0 1 2 3
  CPUs which need to have their frequency coordinated by software: 0
  maximum transition latency: 10.0 us.
  hardware limits: 1.20 GHz - 2.67 GHz
  available frequency steps: 2.67 GHz, 2.67 GHz, 2.53 GHz, 2.40 GHz, 2.27 GHz, 2.13 GHz, 2.00 GHz, 1.87 GHz, 1.73 GHz, 1.60 GHz, 1.47 GHz, 1.33 GHz, 1.20 GHz
  available cpufreq governors: userspace, conservative, powersave, ondemand, performance
  current policy: frequency should be within 1.20 GHz and 2.27 GHz.
                  The governor "performance" may decide which speed to use
                  within this range.
  current CPU frequency is 2.27 GHz.
  cpufreq stats: 2.67 GHz:64.78%, 2.67 GHz:0.81%, 2.53 GHz:1.03%, 2.40 GHz:0.45%, 2.27 GHz:1.38%, 2.13 GHz:1.25%, 2.00 GHz:0.66%, 1.87 GHz:0.48%, 1.73 GHz:0.43%, 1.60 GHz:1.20%, 1.47 GHz:0.80%, 1.33 GHz:0.77%, 1.20 GHz:25.95%  (450)

cat /sys/devices/system/cpu/cpu0/cpufreq/bios_limit

Code:

2. System under load
In this example, I am running stress but anything can trigger this. Even pasting large amounts of data into a terminal(!):

Code:

stress --cpu 8 --io 4 --vm 2 --vm-bytes 128M --timeout 10s

Relevant lines of cpufreq-info output:

Code:

 cpufrequtils 007: cpufreq-info (C) Dominik Brodowski 2004-2009
Report errors and bugs to cpufreq@vger.kernel.org, please.
analyzing CPU 0:
  driver: acpi-cpufreq
  CPUs which run at the same hardware frequency: 0 1 2 3
  CPUs which need to have their frequency coordinated by software: 0
  maximum transition latency: 10.0 us.
  hardware limits: 1.20 GHz - 2.67 GHz
  available frequency steps: 2.67 GHz, 2.67 GHz, 2.53 GHz, 2.40 GHz, 2.27 GHz, 2.13 GHz, 2.00 GHz, 1.87 GHz, 1.73 GHz, 1.60 GHz, 1.47 GHz, 1.33 GHz, 1.20 GHz
  available cpufreq governors: userspace, conservative, powersave, ondemand, performance
  current policy: frequency should be within 1.20 GHz and 1.20 GHz.
                  The governor "performance" may decide which speed to use
                  within this range.
  current CPU frequency is 1.20 GHz.
  cpufreq stats: 2.67 GHz:60.05%, 2.67 GHz:0.75%, 2.53 GHz:2.21%, 2.40 GHz:0.98%, 2.27 GHz:2.46%, 2.13 GHz:1.18%, 2.00 GHz:0.64%, 1.87 GHz:0.47%, 1.73 GHz:0.42%, 1.60 GHz:1.14%, 1.47 GHz:0.77%, 1.33 GHz:0.73%, 1.20 GHz:28.19%  (462)

cat /sys/devices/system/cpu/cpu0/cpufreq/bios_limit

Code:

TobiSGD · 03-27-2012, 11:23 AM

Normally a system clocks down the CPU under load only in one case: when the CPU overheats. Check your cooling system and the temperatures.

terdon · 03-27-2012, 11:43 AM

Hi, TobiSGD, thanks for the reply. Well, I have checked the temperatures. With a previous kernel I was having problems with the fan control daemon so I always keep an eye on them.

sensors| grep Core:

Code:

Core 0:       +61.0°C  (high = +95.0°C, crit = +105.0°C)
Core 2:       +68.0°C  (high = +95.0°C, crit = +105.0°C)

I launched stress like so:

Code:

stress --cpu 8 --io 4 --vm 2 --vm-bytes 128M --timeout 101s

While stress was running, I monitored temp and freq by:

Code:

while [[ 1 == 1 ]]; do  sensors| grep Core; inxi -F | grep Speeds:; done

The clockspeed started decreasing at around 82 degrees. Far from the "high" value, let alone "critical":

Code:

Core 0:       +80.0°C  (high = +95.0°C, crit = +105.0°C)
Core 2:       +85.0°C  (high = +95.0°C, crit = +105.0°C)
           Clock Speeds: (1) 2666.00 MHz (2) 2667.00 MHz (3) 2667.00 MHz (4) 2667.00 MHz

Core 0:       +76.0°C  (high = +95.0°C, crit = +105.0°C)
Core 2:       +82.0°C  (high = +95.0°C, crit = +105.0°C)
           Clock Speeds: (1) 2399.00 MHz (2) 2399.00 MHz (3) 2399.00 MHz (4) 2533.00 MHz

Core 0:       +76.0°C  (high = +95.0°C, crit = +105.0°C)
Core 2:       +82.0°C  (high = +95.0°C, crit = +105.0°C)
           Clock Speeds: (1) 2266.00 MHz (2) 2266.00 MHz (3) 2266.00 MHz (4) 2399.00 MHz

The speed then keeps decreasing until:

Code:

Core 0:       +68.0°C  (high = +95.0°C, crit = +105.0°C)
Core 2:       +71.0°C  (high = +95.0°C, crit = +105.0°C)
           Clock Speeds: (1) 1333.00 MHz (2) 1333.00 MHz (3) 1333.00 MHz (4) 1333.00 MHz

TobiSGD · 03-27-2012, 12:05 PM

I would consider those temperatures much to high, 60°C in idle-state would really bother me. I would recommend to clean the cooling system and may be replace the thermal paste.

terdon · 03-27-2012, 12:15 PM

Well, yes, that's what I thought. However the temperature range reported shows high at +95.0°. It seems very high to me too but I believe these chips are designed to run much hotter than older models.

In any case, you would appear to be right. It looks like the system lowers the clock speed at a specific threshold. I still don't get why this would be done when the temp is so far from the "high" temperature.

273 · 03-27-2012, 12:22 PM

What temperature does the "Motherboard Sensor" give?
I ask because on my system the threshold for the MB is 70C (and CPU 90C) and the CPU and MB sensors are usually within 10C of one another.

TobiSGD · 03-27-2012, 12:24 PM

That high temperature is only 10° below the value when the system does an emergency shut down. It may be possible that the laptop manufacturer has implemented a lower threshold into the BIOS to protect parts that are near the CPU.

terdon · 03-27-2012, 01:38 PM

Quote:

Originally Posted by 273

What temperature does the "Motherboard Sensor" give?
I ask because on my system the threshold for the MB is 70C (and CPU 90C) and the CPU and MB sensors are usually within 10C of one another.

Dunno, cant get info for that one. Getting sensors to work on my laptop is complicated:

Code:

$ inxi -F | grep Sensors
Sensors:   System Temperatures: cpu: 68.0C mobo: N/A gpu: 0.0:72C

Quote:

Originally Posted by TobiSGD

That high temperature is only 10° below the value when the system does an emergency shut down. It may be possible that the laptop manufacturer has implemented a lower threshold into the BIOS to protect parts that are near the CPU.

Yes, that seems quite likely. Well %&#@! I'll have to open it up and check the cpu I guess.