Acceleration of Ultrasound Wave Propagation using a Cluster of GPUs

GPU, CUDA, distributed systems, ultrasound simulations

The simulation of ultrasound wave propagation through biological tissue has a wide range of practical applications including planning therapeutic ultrasound treatments. The major challenge is to ensure the ultrasound focus is accurately placed at the desired target since the surrounding tissue can significantly distort it. Performing accurate ultrasound simulations, however, requires the simulation code to be able to exploit massive performance of multiple modern GPUs to deliver treatment plans within 24 hours. This poster presents a novel domain decomposition based on the Local Fourier basis that reduces the communication overhead by replacing global all-to-all communications with the local nearest-neighbour communication patterns yet exploiting the benefits of spectral methods. The method was implemented using MPI and CUDA to leverage the potential of NVIDIA GPUs. Solution's performance characteristics was evaluated on the Anselm cluster using up to sixteen of its GPU equipped compute nodes. Our GPU based implementation is up to three times faster than the state of the art CPU based implementation running on the same number of Anselm's nodes. It can be stated that, the larger the domain, the better scaling. Naturally, when the ratio between the payload and halo decreases, more additional work has to be conducted, and at some point, the computation part becomes insignificant. Yet we are still able to keep the communication overhead below 50% even for very fine decomposition.