CuSPSolver is a linear sparse system solver based on CUSP Sparse library for OpenSEES. All the algorithms and the preconditioners supported by CUSP can be used in this solver. It can be invoked by a simple statement, which means it is very easy to use. Due to it's hardware and softawre requirements it is not built into the main OpenSees executable.
Hardware Requirements: GPU: NVIDIA’s GPU’s with the CUDA™ architecture.
- The list of CUDA-supported GPUs：https://developer.nvidia.com/cuda-gpus
Software： OS: Windows XP 32/64-bit or later CUDA 5.5 Production Release CuSP 0.4.0 CuSPSolver package`
1) Download the CUDA Toolkit 5.5 package from the official website of NVIDIA (choose the package which is consistent with your OS ), and install it. NOTE currently we are using CUDA 5.5, even though CUDA is now at release 6.0 (we will upgrade once CuSP upgrades to 6.0)
2) Install the CuSP library
3) Download one of the .dll's we provide based on either your <a href="http://opensees.berkeley.edu/OpenSees/code/cusp/32/CuSPSolver.dll"> 32</a> or <a href="http://opensees.berkeley.edu/OpenSees/code/cusp/64/CuSPSolver.dll"> 64 version of windows.
|system CuSP -rTol $RTOL -mInt $MINT -pre $PRE -solver $SOLVER|
|$RTOL||Set the relative tolerance.|
|$MINT||Set the maximum number of iterations.|
|$PRE||Set the preconditioner. can be none, diagonal, and ainv|
|$SOLVER||Set the iterative solver. can be bicg, bicgstab, cg, and gmres.|