The K computer is the world's first supercomputer that broke the 10 petaflops barrier. So how fast is 10 petaflops?
The number ten "peta," or 10 quadrillion corresponds to 1 followed by 16 zeros. In Japanese, this is expressed as one "Kei." That is why this supercomputer is called the K computer.
10 quadrillion worth of computations is equivalent to the world's 7 billion people each performing one computation per second, 24 hours a day for about 17 days. The K computer is able to do all of those computations in just one second.
What made the K computer with its tremendous computational power of 10 petaflops possible?
There is a limit to the computational power and throughput of a single CPU. To scale computational power, a massive number of CPUs have to be interconnected. In the case of the K computer, a CPU with world-class performance and the technology to connect more than 80,000 CPUs were specially developed.
The CPU is the heart of any computer. However, the K computer cannot conduct large-scale calculations at high speed just on the strength of CPU power ? just as we humans cannot act just using our hearts and nothing else. In large-scale computations, various repeating calculation processes are performed, using previous calculation results. Such processes are allocated to multiple CPUs, making data communication between CPUs very important. Since there are 80,000 or more CPUs, a network is needed with the ability to efficiently manage the vast streams of data traffic.
Memory capabilities are also essential in supercomputer design. The vast amounts of data required for the large-scale computations conducted by the K computer at the ultra-high speed of 10 petaflops need to be read in and out of memory. In addition, a storage system is required. This must be capable of holding the vast volumes of computational results that are continuously output.
Fujitsu designed and developed an innovative new technology by combining its know-how of the various building block required to build an ultra-high speed computation machine. This was incorporated in the K computer. The K computer can achieve 10 petaflops because building blocks in the K computer operate at very high efficiency levels in the world.
The K computer consists of a vast number of components including 80,000 or more CPUs and 200,000 or more cables. This makes system resilience a critical success factor. It is essential that
- Individual components are highly fault tolerant
- The system as a whole continues to operate even in the event of partial failure
- Failed components can be replaced while the K computer is in use
The heat generated by the CPUs during operation is managed by a water-cooling system to lower the CPU failure rate.
Another important element - a high-performance, high-reliability unique network to exchange data between the CPUs featuring a 6-dimensional mesh/torus topology was developed. The 6-dimensional mesh/torus topology supports flexible data exchange thanks to a multidimensional CPU interconnection design. Therefore, even if one component fails during operation, the system simply bypasses the failed component and continues to operate.
Share this page