Neural Circuit Simulation of Hodgkin-Huxley Type Neurons Toward Peta Scale Computers

We ported and optimized simulation environment "NEURON" on K computer to simulate a insect brain as multi-compartment Hodgkin-Huxley type model. To use SIMD units of SPARC64VIIIfx (CPU of K computer), we exchanged the order of the compartment loop and the ion channel loop and apply sector caches. These tuning improved single core performance 340 MFLOPS/core to 1560 MFLOPS/core (about 10% efficiency).Spike exchange method of gNEURONh (MPI_Allgather) demands large amount of time in case of 10,000 cores or more and simple asynchronous point-to-point method (MPI_Isend) is not effective either, because of a large number of function calls and long distance of interconnect pathway. To tackle these problems, we adopted MPI/OpenMP hybrid parallelization to reduce interconnect communications and we developed a program to optimize location of neurons on calculation nodes in the 3D torus network. As a these results, we obtained 187 TFLOPS with 196,608 CPU cores.