Yeah, he mentions that as a caveat that all the benchmarks are done using own language scripting abilities... no external functions are called.
And indeed, you're right to highlight that no self-respecting python coder won't use efficient external lapack or arpack libraries to do the computationally heavy matrix operations. numpy is nearly fully C-code under the hood with some C++ as well.
Disappointed/surprised Fortran does significantly worse than C...
Edit: ok, the benchmarks are not only scientific computational ones. For the latter, Fortran still acts very well...
reply