By: 

Sergey Sobolev, CUDA Center of Excellence, Moscow State University

Hosting one of of the most powerful supercomputer centers in the world, Moscow State University is deeply committed to all aspects of supercomputing technology development. As a reflection of this mission, MSU has been a CUDA Center of Excellence (CCOE) for the last four years, using and disseminating GPU technology within Russian science, education and industry.

Currently MSU operates two of the most powerful GPU-based HPC installations in Russia. The flagship system is a new supercomputer “Lomonosov-2,” equipped with 1,536 NVIDIA Tesla K40M GPUs, which enabled it to achieve 2.5 petaflops of peak performance and capture the 22nd spot on the November 2014 TOP500 list. The original “Lomonosov” supercomputer is also still in production, delivering 1.7 petaflops of peak performance, most of which is being delivered by 2,130 NVIDIA X2070/X2090 accelerators. Both supercomputers are equipped with the state-of-the-art GPU-optimized software and are in extremely high user demand for their GPU resources.

MSU is tasked with supporting the entire GPU-based HPC infrastructure, which includes not only providing hardware and software, but also incorporates a significant educational component. In particular, a large number of educational activities on HPC and GPU technologies were provided within the framework of “Supercomputing Education” state projects. The activities were led by MSU and involved more than 600 people.

With the establishment of the CCOE at MSU, a student scholarship contest was launched. Each term five to ten MSU students are awarded scholarships for achieving significant results using GPU computing in their research projects.

An annual “NVIDIA Day at MSU”, has been established at the CCOE providing a unique opportunity for MSU students and researchers to learn more about GPU technologies directly from professionals. From year to year, the event format changes, shifting its focus to different topic areas, such as GPU-intensive applications and GPU programming and code optimization techniques.

The Summer Supercomputing Academy (http://academy.hpc-russia.ru/) is another example of a successful educational collaboration between MSU and NVIDIA. The Academy provides an outstanding two-week event held in June/July, consisting of lectures on topical issues of HPC technologies, advanced training courses (tracks), and master classes, as well as workshops on specific topics. The GPU Programming track covers such topics as CUDA basics, GPU memory types, the CUDA Toolkit, debugging, profiling and optimizing applications, OpenACC and multi-GPU technologies.

MSU is constantly working on building bridges between industry and the research/academic community to improve the use of HPC for commercial purposes. Recently the university initiated a GPU port of the FlowVision CFD package by Tesis LLC. Early results were amazing: performance increased up to 1100x on specific matrix operations. To learn more, please visit the GTC’15 poster section at the upcoming GPU Technology Conference in San Jose, California, on March 17-20.