Äîêóìåíò âçÿò èç êýøà ïîèñêîâîé ìàøèíû. Àäðåñ
îðèãèíàëüíîãî äîêóìåíòà
: http://hpc.msu.ru/?q=node/116
Äàòà èçìåíåíèÿ: Sat Apr 9 22:31:03 2016 Äàòà èíäåêñèðîâàíèÿ: Sat Apr 9 22:31:03 2016 Êîäèðîâêà: IBM-866 |
Web site: http://ccoe.msu.ru (mostly in Russian)
Certification: http://certification.ccoe.msu.ru
As it was defined in 2011, when MSU has been awarded by GPU Center of Excellence (formerly CUDA Center of Excellence) status by NVIDIA, our mission is to create, support and constantly improve a complete GPU HPC infrastructure in MSU and even far beyond. To achieve this, from year to year we implement a wide range of activities.
Accelerating of industry software by using GPU computing facilities is our new and extremely promising trend. We started from FlowVision by Tesis LLC, a CFD simulation software. It is based on the numerical solution of 3D steady and unsteady equations of fluid dynamics and gas. Thanks to a matrix peculiarities used in FlowVision, we achieved following results:
- our block DAXPY kernel is 10x faster comparing to respective cuBLAS procedure;
- our DOT product kernel is 9x faster comparing to respective cuBLAS procedure;
- our LU factorization is 112x faster comparing to previously used CPU-only procedure.
In general, we achieved 50x-1100x acceleration on specific functions using multiple GPUs comparing to old one-core CPU versions.
An online certification for CUDA specialists was launched together with ttgLabs company. Three levels of knowledge are available for certification:
- CUDA API and C/C++ CUDA extensions;
- algorithms realization using CUDA;
- CUDA programs optimization.
Twice a year in the beginning of term GPU CoE MSU announces a scholarship contest for MSU students who use GPU computing in their research projects, i.e. term papers and diploma theses. Main studentsòÀÙ projects areas are biology, molecular dynamics, and materials sciences.
It is a kind of tradition to hold an annual òÀÜNVIDIA Day in MSUòÀÝ, a unique opportunity for MSU students and researches to get more of state-of-the-art GPU technologies directly from NVIDIA professionals. From year to year, the event format is changing. In 2014, it was called òÀÜNVIDIA Technologies Day in MSUòÀÝ. As compared to the previous year events which had been strongly HPC-oriented, the event covered a wide range of popular topics including gaming technologies etc.
Summer Supercomputing Academy is another example of successful collaboration of MSU and NVIDIA on the topic of education. GPU Programming track of Academia covered such topics as CUDA basics, GPU memory types, CUDA Toolkit, debugging, profiling and optimizing of applications, OpenACC and multi-GPU technologies. GPU CoE MSU offers also a special 1-term GPU programming course at the Computational Mathematics and Cybernetics Department of MSU as well as short (2-3 days) for our supercomputer users.