You are not logged in.
I have a bizarre problem on Jessie/MATE (although I saw this on Jessie/XFCE too). As a web developer, I'm in Chromium devtools a lot, inspecting elements. But recently - in the last month or so, whenever I start hovering my mouse pointer over the element nodes in devtools, my CPU clock speed drops to the minimum (~800MHz). As soon as I stop, it jumps back up to between 2-3GHz.
For reference, I'm on a ThinkPad T420S (2011) w/Intel core i7 2640m cpu (2 cores/4 threads - 2.8GHz base clock, 800Mhz min and 3.5GHz turbo). Jessie amd64 is installed.
I've read a related thread here on d1g about CPU performance and clock speed, and tried using the cpupower tools recommended there - but with little to no effect.
My system is using the intel_pstate governor (via cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_driver) and the problem persists with either performance or powersave profiles enabled (selected through the MATE CPU Freq Scaling Monitor panel widget).
I can see in System Monitor that multi-thread load ramps up as I mouse over the devtools inspector's elements, which leads me to suspect that as soon as some multi-thread load is present, the clock speed is minimised. This seems like an overly aggressive power saving strategy to me, as the CPU can't be anywhere near thermal limits. It's also a complete PITA that all the processing power drops away when I need it most!
I have laptop-mode-tools installed and the service enabled, but when I check
service laptop-mode status
I am told
enabled, not active [unchanged]
This is expected as the machine is plugged into AC.
I feel like I'm going mad and have hesitated to post this for the past couple of days, but I could really do with a fresh perspective! Is there a systematic way to work out why this happens on 'reasonable' multi-threaded load (around 40-50% across all 4 threads), or more importantly, how I can stop this throttling?
Thanks...
Offline
Haven't had to deal with this myself yet but from a quick look it seems the intel_pstate driver is kinda special. The way i read https://www.kernel.org/doc/html/v4.12/a … state.html is that at least while in active mode it acts pretty much autonomous. Sorry for just dumping a bunch of links on you but maybe what is described as passive mode at https://www.kernel.org/doc/html/v4.12/a … ssive-mode is interesting in your situation since it seems to give control back to cpufreq as described at https://wiki.debian.org/HowTo/CpuFrequencyScaling.
Offline