high performance computing on graphics processing units: hgpu.org

As we are in the development phase of our own super computer, we have identified several applications which are highly computationally intensive applications for a normal desktop computer to achieve the solution. These identified applications are related to multidisciplinary like bio-medical, mathematics, fluid dynamics, genetic algorithms. We are actually identifying the parallel computations involved in each application which are independent to each other and optimizing it to perform better in our design of supercomputer. We basically use clustered node approach in which one master node distributes the jobs to multiple client nodes after identifying parallel computations involved in the application which are independent to each other. The application uses massively computationally intensive huge data which are SIMD based. We have around 20 nodes with 2 GPUs per node to optimize the computation. We have developed a cluster with job scheduler code to distribute the task to multiple clients after identifying parallel computations which are independent. This paper gives us a great satisfaction in studying the concept of high-performance in various fields of computationally intensive applications. The main application we focussed on is retinal pattern matching using shape extractors using the fast computing method of extracting bifurcation points.