Programmable GPUs: New General Computing Resources Available for Desktop Grids

Researchers have proved that the general computing capability of the mainstream programmable GPU (graph processing unit, the core of graphic card) is more powerful than that of the popular CPU (central processing unit) recently. In this paper, we attempt to develop such a desktop grid, in which the programmable GPUs' powerful general computing ability could be linked in. The hierarchical topology and benchmarks based scheduling algorithm are adopted in the grid to improve the efficiency of the desktop grid. We contrast the performances of GPUs general computing enabled and non-enabled desktop grid to prove the superiority of GPUs' linking supported by some simulation experiments