Detect regular refreshes that memory controller issues every (7.8 us) - in most systems, on normal workloads.
-
Check the image
-
Looking at the first spike, it occured at around ~647,000 (clock cycles). Looking at the second spike, it occurred at around ~662,000 (clock cycles). ~662,000 - ~647,000 = ~15,000. Frequency of the core that this code ran is set to 2.0 GhZ. ~15,000 cc / 2 = ~7,500 ns = ~7,5 us.
-
Before running the code make sure to run the script that sets all of the core's frequencies to their maximum hardware limit (choose "performance" governor). If you want to set a custom value, you need to choose "userspace" governor. However, in some systems, this governor is not supported. "userspace" governor manipulates the frequency by writing to "scaling-setspeed" in /sys/devices/system/cpu/cpu0/cpufreq/. If "$ cat /sys/devices/system/cpu/cpu0/cpufreq/scaling-setpeed" returns "unsupported" go with the "performance" governor. After running the script, make sure it works as expected by using "$ cpufreq-info"
