For me, running HWbot Prime is faster on 2nd run and then drops on 3rd while it is fastest on 4th or 5th run. I tried between different systems but behavior was almost the same everytime across them. UCBench is like running a specific combination of threads. Running i.e. 64 threads alone doesnt achieve the same results. Especially on single cores there seems to be a trick getting a ridicules high result which looks very bugged.