K computer
0 likes54 views
The K computer is a supercomputer developed by Fujitsu for the Japanese Ministry of Education, Culture, Sports, Science and Technology. It has over 80,000 processing nodes, a total of 10 petaflops of processing power, and 30 petabytes of storage. The K computer is installed at the RIKEN research institute in Kobe, Japan and uses a 6-dimensional torus network along with the Tofu interconnect and Message Passing Interface for parallel processing and communication between nodes.
![Node
• CPU: SPARC64™ VIIIfx 8C, 2.0GHz, 8 cores, L2 cache: 6MB,45nm. Low-Power [water
cooled], Fast, Reliable.
• Main memory: 16GB, bandwidth: 64GB/s.
• Network: Tofu interconnect, 6-dimensional mesh/torus topology. ICC (interconnect controller) is a
chip that controls the network.
• Individual components are highly fault tolerant.
• Operation speed: 128GFLOPS.
• Computational nodes: 82,944 nodes (10.62PFLOPS).
• Storage: 30PB](https://image.slidesharecdn.com/kcomputer-180820014821/85/K-computer-7-320.jpg)
![References
• K computer [Fujitsu]
• The Development of the Next-Generation
Supercomputer [RIKEN]
• K computer [HPCI]
• K computer: How to use [RIKEN]
• Top500 / Green500](https://image.slidesharecdn.com/kcomputer-180820014821/85/K-computer-16-320.jpg)
K computer
- 2. K computer • Part of the HPC Infrastructure Initiative led by the Japanese Ministry of Education, Culture, Sports, Science and Technology (MEXT). • RIKEN (Japan’s research institute) chose Fujitsu to develop a powerful supercomputer. • “Design, build, and set up the Next-Generation Supercomputer, the world’s fastest and most advanced computer, with a speed of 10 petaflops.”. 10 peta (10E16) = one “Kei” in Japanese. • Location: K computer is installed at the RIKEN which is located on Port Island, Kobe.
- 6. K computer
- 7. Node • CPU: SPARC64™ VIIIfx 8C, 2.0GHz, 8 cores, L2 cache: 6MB,45nm. Low-Power [water cooled], Fast, Reliable. • Main memory: 16GB, bandwidth: 64GB/s. • Network: Tofu interconnect, 6-dimensional mesh/torus topology. ICC (interconnect controller) is a chip that controls the network. • Individual components are highly fault tolerant. • Operation speed: 128GFLOPS. • Computational nodes: 82,944 nodes (10.62PFLOPS). • Storage: 30PB
- 9. Software • Linux OS providing high portability. • Fortran, XPFortran, C/C++. • MPI 3.1 (Message Passing Interface) Library used in advanced data communication for parallelization. • High-performance, highly functional and system-optimized scientific and numerical library (BLAS, LAPACK, FFTW, OpenMPI).
- 11. Taxonomy (Flynn) • MIMD (Multiple Instructions / Multiple Data) • Multicomputer (multiple homogenous nodes interconnected) • NORMA (non-remote memory access)
- 12. Taxonomy (Comercial) • MPP (Massively Parallel Processors) • Multiple local memories, message passing, high speed interconnected network, high scalable, high cost, hard to program(?).
- 13. Taxonomy (Johnson) • DMMP (Distributed Memory, Message Passing). • “…include both hypercubes such as the NCUBE or FPS T-serie machines and mesh-connected machines such as the AMETEK 2010.”
- 14. Taxonomy (Dongarra/Sterling/Simon/ Strohmaier) • Clustering: Commodity cluster. • Name space: Distributed name space (“variables of one node are not visible directly to another”). • Parallelism: Communicating sequential processes/message passing. • Latency: Caches(?). • c/d/c/c.
- 16. References • K computer [Fujitsu] • The Development of the Next-Generation Supercomputer [RIKEN] • K computer [HPCI] • K computer: How to use [RIKEN] • Top500 / Green500