When is a supercomputer not a supercomputer?

A massive leap forward for Linux and 64bit cluster supercomputing, and a significant cut in high end costs, is how high end workstations firm SGI describes its launch of its latest Linux servers and super-clusters. Aimed at the technical and engineering R&D markets – aerospace and energy, oil and gas – where CAD/CAM and fluid and finite element analysis are key, its new Altix 3000 family is novel. It’s the world’s first to provide for global shared memory across nodes and operating systems; it has multiple platform capability; yet it’s built on standards indlueing SGI’s existing NumaLink hardware and 64bit Intel Itanium 2 processors. In fact, with their shared stoarage, the systems work rather like supercomputers, and the result is new computing performance records way beyond traditional Linux-based clusters and Unix-based servers. SGI notes that the marriage of global shared memory and Linux “creates amazing breakthrough opportunities for technical users on a standards-based platform.” In fact, each node runs a single Linux instance, with up to 64 processors and 512GB of memory. With multiple nodes using SGI’s built-in cluster interconnect, data is transmitted up to 200 times faster than conventional clustering methods, so that the system can scale to hundreds and eventually even thousands of processors, says the company. So it’s not only hugely capable; it’s also extremely scaleable and with its choice of architecture, takes standardisation of 64bit computing to the levels long since achieved with 32bit. SGI argues that it makes top end supercomputer power far more accessible. “It’s a stunning increase in performance,” says John Fleming, high performance computing solutions manager for SGI. And he expects the standardisation to mean that independent software vendors will routinely qualify their applications to the system.