FPGA Acceleration of RankBoost in Web Search Engines

FPGA Acceleration of RankBoost in Web Search Engines is a Microsoft paper about accellerating RankBoost learning algorithm. RankBoost is a variant of AdaBoost, where the weak learners are based on features extracted from the web documents. Microsoft claims of using more that 646 features. Each feature acts as a binary selector and gives a positive vote, if its value is above a threshold. Two techiques are used for speeding up the learning computation:

• a binning and histogram based approximation, for the sequential version of the algorithm.
  
• a FGPA-based implementation, for the parallel version of the algorithm.

I enjoyed reading the paper a lot. Just a question, why the authors are not leveraging off-the-shelf GPU processors, such as CUDA nvidia or PSP3?