Parallel Programming and Different Algorithms based on Step N-1 Results

Hopefully I have selected the correct forum and this note is not too long.

Background: I bought the Jetson TK1 Development Kit to become familiar with the TK1 and GPU computing. My intent is to (eventually) introduce vision processing to my Jr Hi and High school robot club. I was attracted to the TK1 when I ran across the Anaconda Python CUDA compiler.

As a newbie to parallel programming I am lending my TK1 to my “big boy” computer club so they can help me get started. One member has seven computers running 24/7 to gather and process data for his stock market system. I thought his compute problem would be a good fit. He doesn’t think so.

Rational: While there is a lot of array processing in his system, a major sticking point, to him, is the fact that some N+1 steps don’t depend on just the results of step N. The results of step N are used to choose which algorithm(s) will be used in step N+1 calculations.

Real Question: Does the algorithm choice situation stop parallel processing in its tracks?