I performed some benchmarks of simulating a complex AR(1) process. See this 
 ipython notebook 
<http://nbviewer.ipython.org/gist/nbren12/d8fec9a9cdeb917258d0>. Julia is 
about 10x slower than the fortran code, and the fortran is only a few more 
lines of code. Python is my go-to language (no pun indented), and I could 
just use f2py to access this 10x speedup. Assuming I only care about 
julia's performance, and not about its other language features (ie. 
multiple dispatch, macros, etc), I could live with 2-5x slowdown compared 
to fortran. However, if I can expect these sorts of codes to run 10x slower 
than fortran, I think I would prefer  "python +fortran+ f2py" over julia. 

I have a few questions:

1) Am i doing something stupid
2) Are all Monte Carlo type codes susceptible to this slowdown?
3) Would the gap between julia and fortran narrow for higher dimensional 
processes? 

In general, I see the benefits of using a single language over many for a 
single project, and I am very excited about julia, but a 10x slowdown is a 
bit hard to stomach in these sorts of applications.

Reply via email to