From geekbench over cheating code#

> For Linux, the required code can be seen e.g. > that prohibits access, when root/admin rights would be required. > the CPU cycles with a performance counter, unless there is a security setting > On these processors it is easy to measure the average CPU frequency, by counting > hundreds or thousands of Geekbench results exist, but all are garbage because they are not comparable. > Anyway, these frequency values are interesting especially > the benchmark should not record the average clock frequency, but where possible, it should be recorded. > where access to the performance counters is prohibited, > On a processor without performance counters or on a system > not worthwhile, because they must be calibrated for each processor model) timing integer divisions or other long instructions are > To obtain the number of CPU cycles, there is no other good way except using one performance > In all modern processors TSC is incremented at a constant frequency, regardless what the processor cores do. > The link from Stackoverflow is obsolete. Dividing the 2 numbers provides the average frequency. > of CPU cycles from test start to test end. > The duration of the test is already measured by any benchmark, you just need one extra number, the count > Where it is possible, it is not difficult. > Some good links there, but it is easier on Linux than Windows (big surprise), but it is possible. > I agree, but this is a difficult problem because of turbo/thermal throttling, etc. > systems, where the overclocking problem exists, it could be done using one performance counter.

> possible on all platforms, but at least on Intel/AMD > It would be great if the Geekbench tests would also measure the average clock frequency during the test

From geekbench over cheating