We investigate the opportunities for using GPUs to perform numerical integration for finite
element simulations. The degree of concurrency and memory usage patterns are analyzed
for different types of finite element approximations. The results of numerical experiments
designed to test execution efficiency on GPUs are presented. We draw some conclusions
concerning advantages and disadvantages of off-loading numerical integration to GPUs for
finite element calculations.

The algorithmic and computational aspects of 3D finite element numerical integration on
GPUs are investigated in the paper. The special stress is put on selecting the proper
parallelization strategies depending upon the properties of FEM problems solved and
approximations used. The close interplay between the available computational resources of
GPUs and the possible implementation strategies and obtained results is observed.