CUDA Accelerates Computational Discovery of New Nanoporous Materials

While GPUs were originally developed for computer graphics, they are now being used by scientists to help solve important engineering problems. The performance gains from parallelizing molecular simulation codes in CUDA have facilitated efforts to computationally evaluate large databases of nanoporous material structures for several applications.

Researchers from UC Berkeley and Lawrence Berkeley National Laboratory discuss how CUDA has facilitated their materials research.

Materials Discovery
A nanoporous material can be abstracted as a raveled-up surface. On the left is the unit cell of the IRMOF-1 crystal structure. On the right is a depiction of the surface that IRMOF-1 forms.

Molecular simulations are computationally intensive, and the number of material structures that we are testing in silico is rapidly expanding. In their recent work, they screened a database of over half a million nanoporous material structures for natural gas storage using molecular simulations. For simulating gas adsorption in such a large number of materials, fast, parallelized computer code written in CUDA to run on GPUs accelerates our research progress dramatically.

Using a Tesla K40 GPU for their CUDA code, the throughput is 40 times that of the OpenMP-parallelized code!

Read the entire blog on Parallel Forall >>