The Tianhe-2 was delivered in 2013, It has a heterogeneous accelerated architecture consisting of CPU with MIC accelerators. With 33,9 Petaflop/s of Linpack performance, it was 6 times in a row the fastest supercomputer in the world. It is installed at the National Supercomputing Center in Guangzhou and will be upgraded to 100 Petaflop/s later this year.
The Sunway Taihulight was delivered in 2016 and is implemented too with home-grown Chinese Shenwei many-core processors. In total, the system has 10 million cores. The peak performance is 125 Petaflop/s. The Linpack performance is 93 Petflop/s. It is the fastest supercomputer in the world since June 2016. The system is installed at the National Supercomputing Centre in Wuxi.
As part of CNGrid there are now 14 connected supercomputing sites in China, and 3 new ones.
In HPC applications there are now many areas covered, including aircraft design, high-speed train design, oil and gas exploration, drug discovery, ensemble weather forecasting, bio-information, car development and more.
There is now a new key research project on HPC. HPC has been identified as a priority subject under the key R&D programme. There have been strategic studies performed since 2013 and an HPC proposal for the 135th 5-year plan was submitted in early 2015. After serious evaluation it was approved in October 2015. The project was formally launched in early 2016.
The major tasks of this project include the development of exascale supercomputers; develop HPC applications; and create an HPC ecosystem.The exascale system should be delivered by the end of 2020. It will be based on home-grown Chinese processors. The technology developed will be transferred to high-end servers.
A first call for proposals was issued in February 2016. 19 projects passed evaluation and were started in July 2016. A second call for proposals was issued in October 2016. Proposals are now under evaluation. A third call - the 2018 call - will be issued soon. This third call will focus on exascale computing development.
In three projects, there are activities to develop three prototypes for future exascale systems. The possibilities to implement these at exascale level are also explored. The three prototypes are developed by the Sunway, Tianhe, and Sugon teams. Different kinds of architectures are explored, including accelerated architectures, special purpose architectures, many-core, and x86 processor based ones. Several new technologies will be developed. Energy efficiency is the biggest obstacle and will be an important factor influencing design choices.
The team that will be developing the real exascale supercomputing system will be chosen from these three teams that are building the prototypes.
Some numerical systems for specific application domains will also be developed. This includes a numerical aircraft, a numerical earth, a numerical reactor, and a numerical engine. The first two have been launched as first round projects.
A parallel computing framework for exascale computing will also be developed. It includes structured and unstructured mesh solvers, finite element solvers and graph computing. At least 40 software packages will be developed that have million-core parallelism.
As part of the HPC environment CNGrid will be upgraded to over 500 petaflop/s in computing resources, over 500 Petabytes of storage, over 500 application tools and software packages, and over 5000 users.
Applications villages based on the national HPC environment will be developed for industrial product design and drug discovery/bio-informatics.
For more details on the new Chinese exascale efforts see China's effort on HPC in the next 5 years - from exascale prototypes to exascale system , which covers the longer presentation Depei Qian provided, later on the same day of the HPC in Asia workshop.
Recordings of these two presentations are available on Primeur's video channel.