During June 23-26, 2015, a summer school on HPC organized by Algorithms Division, CSRC has been held in CSRC. The summer school was attended by over 50 participants, with the majority coming from CSRC faculty, postdoctoral fellows, and graduate students. The school included lectures and hand-on workshop, given by IBM research scientist Dr. Leopold Grinberg and Yuhang Tang from Applied Mathematics, Brown University, USA.
In the first two days (6/23-24) of the summer school, Dr. Grinberg introduced the basic structure of computation nodes in modern parallel computer system in both shared and distributed memory platforms and their difference and special features, discussed OpenMP and MPI programming language and their implementations in various scientific computing examples. In the next two days (June 25-26), Yuhang presented the GPU structure and its special capabilities, CUDA language and relevant application libraries: CuBlas, CuSparse, THRUST, the layered memory structure of GPU, and core optimization and the strategy of using CUDA to conduct computation with GPUs. Their lectures have attracted much attention from the participants and produced lively interactions among the attendees and lecturers.
A special feature of this summer school was the hand-on sessions on a newly purchased CPU/GPU workstation (4 K40 GPU cards) by the center for the participants. In each afternoon of the summer school, the lecturer would provide specific example codes to be practiced by the students, who gained a deeper understanding of the theoretical issues and algorithms discussed in the lectures in terms of realistic application problems.
The summer school adjourned in the afternoon of June 26 and achieved the expected objectives of the organizer and the students have mastered the basic skills in parallel computing with OpenMP and MPI and CUDA programming in GPU machines, which will prepare them well for carrying out large-scale computational scientific research.