Accelerating incompressible flow computations with a Pthreads-CUDA implementation on small-footprint multi-GPU platforms