Microway has announced its most powerful Tesla GPU based platform to date. The OctoPuter compute engine includes 2x quad-core Intel Nehalem or 2x six-core AMD Opteron processors. The 8 PCI-Express Gen 2.0 x16 slots support up to eight Tesla C1060s, each with 4 GB in a chassis designed for multi-GPU applications. Large memory capacity (up to 144 GB) and a high efficiency, 2+1 redundant power supply complete the package. This platform is suited to HPC applications and virtualization. This configuration can be used as a building block with high-speed InfiniBand interconnect for dense GPU/CPU computing. “This high density solution is the ideal platform for HPC users with coarse grained applications that run well in a 1 TFLOP environment. These include ray tracing, Monte Carlo simulations, DSP applications, and rendering. OctoPuter is the ideal platform for building a cost-effective rendering farm,” said Stephen Fried, Microway’s president and CTO. The Tech-X demo Microway is showing at booth 434 at SC ’09 is a comparison of a numerical simulation run purely on the CPU and with GPU acceleration. The entire application is written in the high-level language IDL by ITT VIS. The port to GPUs is using Tech-X GPULib and the multi-GPU parallelization was accomplished via Tech-X mpiDL (MPI bindings for IDL). The FDTD algorithm is used for modeling the propagation of electromagnetic waves. This algorithm solves the full set of Maxwell's equations on a regular grid, which is 2000x1000 cells large in this case. FDTD is used for a range of applications, from cell-phone simulations to laser-matter interaction. The rate of grid cells computed per second are plotted on the graphs; in this case there are about 5 million cells per CPU and 400 million cells per GPU. The more GPUs you have, the faster the overall computation runs. Tech-X will also exhibit this demo in Booth 1713. For more information, visit Microway. Sources: Press materials received from the company and additional information gleaned from the company's website.
|
