Hi, This information might be interesting to some of you. I have done some profiling in Overflow in order to measure the overhead of spliting the process into nodes (the overhead of Overflow vs. C/C++). The tests were performed on my Athlon 500. Here are the timings I got (+- 10%): Node overhead: 2.7 us (1350 clock cycles) Loop overhead: 1.1 us (550 clock cycles) The node overhead is the time it takes to execute the simplest node. It means that you should try, on average, to have nodes that do more that 2.7 us of calculation. Loop overhead is the minimum time for one loop iteration. Basically, it means that the loop overhead is negligeable. I think these timings are quite reasonable. What that means is that if you use Overflow (or Piper, of course!) to compute FFT's, then your everhead is negligeable, but it you use it for elementary operations (+,-,*,/) at the node level, then expect to pay a high price. When I tried the same with matlab I got: Function (node) overhead: 8-18 us (8 for a builtin, 18 for an m-file) Loop overhead: < 1 us (I cannot measure more precisly) Jean-Marc -- Jean-Marc Valin Universite de Sherbrooke - Genie Electrique valj01 at gel.usherb.ca