1 \section{Conclusion}\label{ch6:conclu}
3 In this chapter, different methodologies that effectively take advantage of
4 current cluster technologies with GPUs have been presented. Beyond the simple
5 collaboration of several nodes that include GPUs, we have addressed parallel
6 schemes to efficiently overlap communications with computations (including GPU
7 transfers), and also computations on the CPU with computations on the
8 GPU. Moreover, parallel schemes for synchronous as well as asynchronous
9 iterative processes have been proposed. The proposed schemes have been validated
10 experimentally and provide the expected behavior. Finally, as a prospect we have
11 developed a programming tool that will, in middle or long term, provide all the
12 required technical elements to implement the proposed schemes in a single tool,
13 without requiring several external libraries.
15 We conclude that GPU computations are very well suited to achieve overlap with
16 CPU computations and communications and they can be fully integrated in
17 algorithmic schemes combining several levels of parallelism.
22 %%% ispell-dictionary: "american"
24 %%% TeX-master: "../../BookGPU.tex"
