This project is for the acquisition of a GPU cluster dedicated to research on High Performance Computing. The selected infrastructure (a cluster of GPUs) presents itself as a serious candidate for the future of High Performance Computing. Indeed, the theoretical cost efficiency(GFlop/sec/dollar) and the theoretical power efficiency (GFlop/sec/Watt) scores very high compared to current other high performance computing platforms. However, two huge challenges are whether relevant application codes can reach a reasonable fraction of the peak of these computing platforms and how to reach this performance level. The PIs collaborative effort answers these research questions from various angles. The proposed infrastructure is supporting their research on these issues.
The codes produced by this collaborative effort are included in well-known software packages (LAPACK, NVIDIA CUBLAS, ScaLAPACK, WRF).