Modern Graphics Processing Units (GPUs) give spectacular performance by using parallel execution in a constrained architecture designed for real time generation and rendering of three dimensional scenes. A major obstacle to their efficient use with traditional tree genetic programming has been each parallel path way executes the same program (a SIMD approach).
This SIMD program interpreter runs in parallel allowing huge GP populations to be evaluated on the GPU.
I will show: time series prediction and bioinformatics (protein predication) examples using populations of up to a million programs evolving for up to a thousand generations on which the GPU interpreter gives sustained performance of almost 1 billion GP operations per second on a single NVIDIA GeForce 8800 GTX.