For both band-structure computations and finite element simulations we have a dedicated cluster housed at the IIT tech park. The cluster includes one head node and 4 compute nodes. Each compute node has a mixed CPU/GPU architecture with two CPUs and two GPUs. We are currently using NVIDIA’s GTX-780 graphics cards in conjunction with Intel Xeon processors.
As far as possible we try to use open source tools for research. These include the following:
- Python with ATLAS BLAS/LAPACK, numpy, scipy, matplotbib, mpi4py and other extensions serve as equivalent to parallel MATLAB.
- We use pycuda in conjunction with python for computations on the GPU.
- GMSH is used for creation of geometries and mesh generation.
- FEniCS is used for solving the weak variational forms which lie at the heart of Finite Elements.
- Paraview is used for visualization.
We have custom CPU/GPU implementations of various variational methods used for solving the phononic band-structure. These include the Rayleigh quotient, inverse Rayleigh quotient, the mixed-quotient, and PWE. Some of these converge faster for high contrast unit cells and some are appropriate for mixed solid-fluid unit cells. All these are implemented in Python and use the software mentioned above.
We also use a combination of COMSOL and MATLAB to solve problems wherever solutions in FEniCS and Python are hard to implement.
Here is an article on how to access to the cluster from on and off campus locations.