Computation Demand Drives Computer Design

The growing use of computational fluid dynamics (CFD) and other highly compute-intensive engineering-software applications is driving innovation in the hardware field. Among developments: Both SGI and Cray Inc. are pushing new computer systems that provide the speed and power to help engineers run sophisticated analysis programs faster.

Earlier this month, Cray released for commercial sale its new XD1^(TM) Opteron/ Linux-based supercomputer, "purpose-built," the company says, for high-performance computer-aided-engineering (CAE) applications, including finite element analysis (FEA) and CFD.

The base unit provides 12 AMD Opteron 248 processors. The company says the supercomputer runs all x86 Linux applications using the MPI (Message Passing Interface) programming model. The system uses a direct connected processor architecture to directly link processors to each other and memory. That, Cray says, eliminates interconnect bottlenecks and provides 30 times greater bandwidth and 30 times lower latency than typical cluster systems.

Meanwhile, SGI is planning an upgrade to its Altix^(R) line, due out later this year. The Altix 3000 system runs a single Linux OS image with 256 Intel^(R) Itanium^(R) 2 processors and up to 4 Tbyte of memory.

The Altix series uses the company's NUMAflex^(TM) shared-memory approach rather than distributed memory. Shared memory means that a single memory address space is visible to all system resources, including microprocessors and I/O, across all nodes. SGI says that shared memory allows access to all data in the system's memory directly, without going through I/O or networking bottlenecks.

Cray claims that the number of shared-memory applications is declining because of the cost of scalability. SGI counters that NASA chose the Altix as the foundation for Project Columbia, a project expected to fuel scientific breakthroughs in space exploration and aerospace engineering.

In another development, SGI this month released the PRISM^(TM), a Linux-based visualization system. The system will address terabyte-sized, highly complex data as a single image, increasing speed of use.

Says Simon Hayhurst, SGI's product line manager for visualization systems, the PRISM will speed up the design process in several ways. "For one, it will enable a tighter coupling between analysis and design." Engineers could actually do FEA as they design by running a coarse grid to make incremental improvements, he says. Another area where the PRISM could save time: presentations.