Modern CPUs can run at several clock speeds, to choose a balance between electricity usage/heat generation and performance. Modern kernels support various ways of controlling this - the most common out-of-the-box one is 'ondemand'.
The ondemand policy automatically changes the CPU speed while the system is running - unfortunately, there's lag involved, so this makes benchmark results unreliable.
On Linux systems, there should be a directory for each CPU in /sys/devices/system/cpu. If so, and CPU frequency scaling is enable, /sys/devices/system/cpu/cpu/cpufreq/scaling_governor should exist, and contain the name of the current policy.
If that file does not exist, we assume everything is good. If it does, we check that contains 'performance', for every CPU.
The generic solution is to run this as root:
for file in /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor; do
echo performance > $file
doneYour distribution or desktop environment may provide power management features that interfere with this. You will need to check their documentation as appropriate. Also, if you are running a desktop environment, this may indicate that your system is more generally not representative of a production server.