Node architecture The K computer comprised 88,128 2.0 GHz
eight-core SPARC64 VIIIfx processors contained in 864 cabinets, for a total of 705,024 cores, manufactured by Fujitsu with
45 nm CMOS technology. Each cabinet contained 96 computing nodes, in addition to six I/O nodes. Each computing node contained a single processor and 16 GB of memory. The computer's
water cooling system was designed to minimize
failure rate and
power consumption.
Network The nodes were connected by Fujitsu's proprietary
torus fusion (
Tofu) interconnect.
File system The system adopted a two-level local/global
file system with parallel/distributed functions, and provided users with an automatic
staging function for moving files between global and local file systems. Fujitsu developed an optimized parallel file system based on
Lustre, called the Fujitsu Exabyte File System (FEFS), which is scalable to several hundred petabytes.
Power consumption Although the K computer reported the highest total power consumption (9.89
MW the equivalent of almost 10,000 suburban homes) on the June 2011 TOP500 list, it is relatively efficient, achieving 824.6
GFlop/kW. This is 29.8% more efficient than China's NUDT TH MPP (ranked #2 in 2011), and 225.8% more efficient than Oak Ridge's Jaguar-Cray XT5-HE (ranked #3 in 2011). However, K's power efficiency still fell far short of the 2097.2 GFlops/kWatt supercomputer record set by IBM's NNSA/SC Blue Gene/Q Prototype 2. For comparison, the average power consumption of a TOP 10 system in 2011 was 4.3 MW, and the average efficiency was 463.7 GFlop/kW. According to TOP500 compiler
Jack Dongarra, professor of electrical engineering and computer science at the
University of Tennessee, the K computer's performance equaled "one million linked desktop computers". The computer's annual running costs were estimated at US$10 million. ==K Computer Mae rapid transit station==