Continue to get acquainted with the concept supercomputer SGI Molecule, the foundation for the construction of which had modest in terms of performance chips Intel Atom. The basic idea was to offer designers a solution to the challenges that the large number of computing nodes, albeit far from bleschuschih their capabilities, is far more important than the productivity of each site separately. According to the developers, the use of this concept optimized for mobile chips Intel Atom would make an impressive performance energy efficiency - both in absolute terms, total, and in relative terms of productivity watts. module concept Molecule is a system board size of a credit card, which vpayan dual-core Atom N330 Diamondville, operating at a frequency of 1, 6 GHz, and singling out while up to 8 W heat. Strictly speaking, N330 is a two processor, combined in one case. With the advent of the next generation of Atom, known as Lincroft, and equipped with embedded memory controller, it will be possible to further simplify the fee, rather than the more discrete components chipset, as the concept of Molecule no existence separate module of a storage device, other than RAM. DDR2 chips total capacity of 2 GB and a processor unsolder directly on the motherboard. Two separate boards fixed on the floor keramicheskom cartridge, as the carrier chassis, heat sink and air. This unit, a cartridge with the cards, SGI named Kelvin. Connecting with one another, forming system cartridges vozduhootvodyaschih channels. Interconnect for connection to external storage systems and networks are a subject for further development, but so far it seems that they will be built similar to the system bus chassis blade-servers. Concept system, displayed at the exhibition, is a high density rackmount design height 3U, contains 180 cards with chips Atom, that is, in the amount of 360 processor cores. Through the support gipertreydinga system can ensure that the 720 virtual streams of commands at a speed of the exchange memory 720 Gb/s. The developers note that with the correct architecture Interconnect full search of the entire memory system can be implemented in one second. They believe that such a system could be particularly effective in cases where the use of the cache, or the performance of teams change the sequence does not provide the desired result, while support for multiple streams is very effective. As these applications have been identified as a computer gazogidrodinamika, seismic processes and stochastic modeling. SGI said that while that will explore interest in the concept of the system from potential customers, and with sufficient interest will be the creation of a prototype, which may take approximately 12-18 months.