Because Fortran was listed as x.xxx, I figured I'd try it. Taking the minimum of 100 runs of gcc/gfortran, 200 runs of g++, and 4 runs of Julia 1.1-dev: gcc: 3

Because Fortran was listed as x.xxx, I figured I'd try it. Taking the minimum of 100 runs of gcc/gfortran, 200 runs of g++, and 4 runs of Julia 1.1-dev: gcc: 3.4866 g++: 3.4428 gfortran: 3.4953 Julia: 8.0378 I was doing other things while the benchmarks ran, which added noise. For the first run of g++, the mean was slightly higher than C's mean, while the standard deviation was much higher than C or Fortran's. So I reran it for C++, and then decided to just report the minimums for everything instead, because the minimum is probably the least biased measure. So long as there are no allocations that occasionally get cleaned up by a garbage collector. If there were, the GC cost should get amortized over all the runs that contributed to it. We don't have to worry about this here. While the C and C++ assembly for the fib function was the same, Fortran's was different. Julia's assembly was extremely brief in comparison, because it doesn't have any recursion optimizations -- it is just the check and then two calls to itself.