I recently stumbled upon two academic researchers who used CUDA libraries in order to integrate GPU processing with genetic training for algorithms trained on FX markets. I highly recommend taking a look at their methodology as this was previously considered impossible to integrate into genetic algorithms http://people.scs.carleton.ca/~dmckenne/5704/Paper/Final_Paper.pdf
Also wanted to note that just recently, CUDA was used in order to achieve 6000x performance increase on NVIDIA GPUs backtesting financial markets. I don't know if the libraries for CUDA were updated to better handle this type of computation but it has been achieved. There should be a way to introduce GPU computing across the board for different processes in SQX.
"I use Python + CUDA to create a backtesting system, which can output backtesting results for 38,000 different indicator parameters in about 150 seconds. This is much more efficient compared to SQ's CPU utilization."