In recent decades, increasing ultrasound frame rates has been the main motivation behind many novel ultrasound imaging applications [1]-[3]. With this work, we propose an efficient ultrafast FPGA beamformer that applies coherent compounding, through a delay-reuse optimization.

Computational neuroscience aims to investigate and explain the behaviour and functions of neural structures, through mathematical models. Due to the models' complexity, they can only be explored through computer simulation. Modern research in this field is increasingly adopting large networks of neurons, and diverse, physiologically-detailed neuron models, based on the extended Hodgkin-Huxley (eHH) formalism. However, existing eHH simulators either support highly specific neuron models, or they provide low computational performance, making model exploration costly in time and effort. This work introduces a simulator for extended Hodgkin-Huxley neural networks, on multiprocessing platforms. This simulator supports a broad range of neuron models, while still providing high performance. Simulator performance is evaluated against varying neuron complexity parameters, network size and density, and thread-level parallelism. Results indicate performance is within existing literature for single-model eHH codes, and scales well for large CPU core counts. Ultimately, this application combines model flexibility with high performance, and can serve as a new tool in computational neuroscience.

Detailed brain modeling has been presenting significant challenges to the world of high-performance computing (HPC), posing computational problems that can benefit from modern hardware-acceleration technologies. We explore the capacity of GPUs for simulating large-scale neuronal networks based on the Adaptive Exponential neuron-model, which is widely used in the neuroscientific community. Our GPU-powered simulator acts as a benchmark to evaluate the strengths and limitations of modern GPUs, as well as to explore their scaling properties when simulating large neural networks. This work presents an optimized GPU implementation that outperforms a reference multicore implementation by 50x, whereas utilizing a dual-GPU configuration can deliver a speedup of 90x for networks of 20,000 fully interconnected AdEx neurons.