NVIDIA HP joins hands to push GPU computing clusters: Just $100,000 oh pro

NVIDIA HP today announced a limited edition "GPU Starter Kit" to provide researchers with an easy-to-use, pre-configured GPU computing cluster that can be used directly without debugging.

The cluster consists of eight HP ProLiant SL390 G7 servers and contains 24 NVIDIA Tesla M2070 GPUs, 16 Intel Xeon 5600 CPUs, 13.5 TFlops of peak single-precision floating-point performance (13.5 trillion times per second), and is pre-installed. The latest NVIDIA CUDA 4.0 parallel computing software supports C, C++, Fortran programming, and provides a large number of GPU acceleration libraries such as FFT, BLAS, LAPACK, RNG, SPARSE.

There are already several high-performance supercomputers using HP's ProLiant SL390 G7 servers, such as Tsubame 2.0, which is regarded as the most energy-efficient Tokyo Institute of Technology in the world, and the 32-GPU computing performance can exceed 1200 CPUs.

This GPU Computing Starter Kit is priced at $99,900, seemingly astronomical but in fact is already about half the typical offer for this type of product, and at the same time offers a large number of third-party development tools at a discounted price.