Numerical Simulations Using Programmable Graphics Processing Units (GPUs)

The latest GPUs are reported to reach up to 200 billion floating point operations per second and to have price performance of 0.1 cents per Mflop. These facts raise great interest in the plausibility of extending the GPUs' use to non-graphics applications, in particular numerical simulations on structured grids (lattice).

This talk will be a brief introduction to programmable graphics hardware and its use for non-graphics applications. I will review previous work on using GPUs for non-graphics applications, explain the graphics pipeline, and what the programmable GPUs do. I will discuss the specifications and block diagram for one particular graphics card, nVidia's GeForce FX 1000, known as NV30. Also, I will discuss the implementation of two probability based models, namely the Ising and percolation models. The implementations were used in benchmarking the NV30 graphics card and its high level compiler, called Cg.

A general conclusion is that it is feasible to use GPUs for numerical simulations. Preliminary results also show that it is feasible to use them in parallel.

.

Top of Page