Last week French supercomputing specialist Bull announced that it extended and updated its bullx family of supercomputers with the launch of its new bullx supernodes (S-Series) line. According to the company, these four-socket SMP (Symmetric Multi-Processing) server nodes can be expanded to support up to 128 cores (8, 12 or 16 processors) and 2TB of RAM.
Designed in cooperation with the French Atomic Energy Authority, the processing beasts capitalize on the latest generation of Intel Xeon 7500 CPUs (Nehalem-EX), and are intended to be used as part of a larger supercomputing cluster--just think of each supernode as part of a multi-unit rack. "Using these processors, Bull has designed an extendable SMP server capable of meeting the needs of even the most memory-hungry applications," the company said.
When available, Bull will offer two versions in its new S-Series: the ultra-compact high-end bullx S6010 compute node, and the bullx S6030 service node. The S6010, seen in the image to the right, features L-shaped 1.5U drawers--one upside-down on top of the other--to form a 3U drawer that can house a 8 or 16-processor configuration. Bull's other supernode, the S6030 model, provides "advanced connectivity functions, a redundant power supply and extended storage options," making them suitable to serve as management or I/O nodes.
In addition to the overall horsepower, the new S-Series is environmentally friendly, using an ultra-capacitor module to improve the efficiency of the electrical power supply by 10 to 15-percent. The supernodes also use a cooling door that consumes 75-percent less energy than standard air-conditioning, a processor control system that optimizes energy consumption, and an infrastructure design that optimizes the Flops/m² ratio.
...not that I would be able to afford it anyways.
What program would be able to use all of it?
What exactly does the supercap do to make the power supply more efficient? Just smooth out spikes and dips? Supply the high startup current to get the fans running?
Usually, when dealing with physical processes, like what happens with a nuclear weapon, you'd use what's known as Monte Carlo codes. Monte Carlo codes model highly random sitations (thus the name Monte Carlo cause you're rolling a die to see what happens effectively) like neutron scatter/absorbtion through a medium (in this case a neutron through say Uranium or Lithium Hydride). In essence, it's a way to test how powerful a nuclear weapon can be without actually setting it off (or even to look at how air travels over a foil like that of a wing). Because you are modelling such a complex system, imagine having to follow trillions of particles over a very short period of time (like nanoseconds in the case of a nuclear weapon), the programs can get quite involved. So when you're getting to situations that in depth, it takes a lot computing power.