Ultrasound simulation on a GPGPU - Reducing memory bottleneck using compression

The simulation of ultrasound signals is a computationally intensive task from the perspective of the number of floating point operations required and memory bandwidth. This project involves simulating realistic ultrasound signals using the Westervelt Equation on a GPGPU with a focus on the memory bandwidth bottleneck. This project explores a data compression technique which uses floating point to fixed point conversion to reduce the memory bandwidth. Using this approach, we are able to achieve a performance gain of around 20% with respect to running time while reducing the memory bandwidth by around 50% over the baseline GPU implementation for grid sizes of 2560 x 2560 and above.