Through GPU programming it is possible to get results faster at the same cost or same performance with less costs. In practice, for the trading institution this translates to lower latency, opens new possibilities for new compute-expensive models to be implemented, or saves costs. Recent developments made GPU programming easier than how was in the past. During the presentation we describe the usual steps we take when porting an existing pricing model on the GPU, both on C++ and Python, showing the tools currently available and the potential speed-up achievable.