The accuracy of CPU performance benchmarks

Measuring how powerful a CPU is can be a very tricky job.

There are benchmarking websites out there that give you a feel of a CPU capabilities, but at the end of the day it is hard to talk about absolute performance power from a CPU. As with most things in life YMMV (your mileage may vary). Which means that how good a CPU is, is related to what the CPU is doing.

You have CPU frequencies, cache type and sizes, memory bandwidth and so on. If you just want to crunch numbers, the memory bandwidth wont do you any good. If you want to reencode a 4K movie, well you have to move in and out that entire movie, frame by frame from the memory to the CPU at some point, so memory lines and speeds are critical.

At the end of the day, you need to run your workload into a CPU/system so you can see how well it performs.

My destop: Intel i7 5820K (stock clock is 3.3Ghz but I have it overclocked on a water block to 4.0Ghz).
My laptop: Intel i7 3631QM

Im working on some number crunching code and here is my performance table:

1core laptop desktop d/l
cpu_load 3968 8264 2.08
cpu_mips 6229 15211 2.19

As we can see in the table, the CPU benchmarks on the web say the desktop is 2.19 faster than the laptop (this is on a per-core basis).

The real number measured giving the CPU a real workload is 2.08 which is very close actually.

cpu_load one_core all_cores d/l
laptop 3968 8 31744
deskptop 8264 12 99168

So when using all cores the desktop can crunch numbers 3.12 times faster.