Depends. In the first it will depend on the branch predictor which will depend on the relative expected magnitudes of num and x In the 2nd, which I assume shou

Depends. In the first it will depend on the branch predictor which will depend on the relative expected magnitudes of num and x In the 2nd, which I assume should be { return num*num > x * x; } then it depends on the micro-arch, as it's one basic block so no branches and assuming a deep pipeline on x64, one multiplier (pipelined), probably this is faster for 'random-ish' num and x.