high performance computing on graphics processing units: hgpu.org

The multi-scale nature of the electrophysiology problem requires the use of fine temporal and spatial resolutions leading to models with millions of degrees of freedom that need to be solved for a thousand time steps. Solution of this problem requires the use of algorithms with higher level of parallelism in multi-core platforms. The newer programmable graphic processing units (GPU) has become a highly parallel, multithreaded, many-core processor with tremendous computational horsepower. This paper presents results obtained using HESIC, a novel electrophysiology simulation software entirely developed in CUDA. The software implements implicit and explicit solvers for the monodomain model, using operator splitting. The ten Tussher and Panfilov cell model (TP06) has been considered in isolated single cell, and one-, two- and three-dimension anisotropic tissue models discretized with structured meshes with 0.1mm resolution have been used for benchmarking. Results obtained with a NVIDIA C2090 GPU on simulating 100ms of cell activity show that GPU performs non-linearly with the number of degrees of freedom. For small problems (<1000 nodes) the GPU underperforms a single CPU due to GPU’s lower clock speed. However, as the problem size increases, the GPU outperforms a single CPU up to 180 fold for the integration of the ionic model, and up to 70 fold for the three-dimensional tissue model.These results points GPU computing as a promising and economic alternative for high performance simulations of heart electrophysiology.