HP delivers GPU Starter Kit with 24 NVIDIA Tesla M2070 GPUs

NVIDIA today announced the availability of a limited-edition GPU Starter Kit from HP, a pre-configured system that provides researchers with a ready-to-use GPU computing cluster, straight out of the box. It consists of eight HP ProLiant SL390 G7 servers, containing 24 NVIDIA(R) Tesla(TM) M2070 GPUs, 16 CPUs, and is pre-configured with the latest NVIDIA CUDA(R) 4.0 parallel computing software.

Designed for ease of deployment and to address the growing availability of major GPU-enabled scientific applications, the GPU Starter Kit enables every university, government research department and enterprise customer to quickly deploy a rack of GPU-accelerated servers. The system delivers 13.5 teraflops of peak performance and is priced at $99,000, roughly 50 percent below the typical list price. The kit also includes a broad set of third-party development tools at discounted prices to facilitate application optimization.

"Growing demand for GPU computing has fueled the need for a fully integrated, robust and affordable development platform that enables developers to easily create new, accelerated applications," said Sumit Gupta, manager of Tesla products at NVIDIA. "As a result, NVIDIA and HP developed the GPU Starter Kit to remove some of the last hurdles to the mass adoption of GPU computing -- namely, the cost and time of system implementation."

HP ProLiant SL390 G7 servers equipped with Tesla GPUs are the computing platform of choice for leading supercomputers, such as the world's greenest production supercomputer(i), Tokyo Institute of Technology's Tsubame 2.0. They also power the Keeneland system -- managed by Georgia Tech, Oak Ridge National Lab and the National Institute for Computational Sciences -- enabling researchers and students to advance scientific discovery in various disciplines. Some of these efforts include:

To better understand protein-DNA interactions with applications in future drug design, researchers Bo Hong, at Georgia Tech, and Jun-tao Guo, at University of North Carolina, Charlotte, are relying on GPUs to accelerate protein-DNA docking simulations. With 32 GPUs in the Keeneland cluster, they are able to achieve higher performance than with 1,200 CPUs alone.

In search of advanced materials for wide ranging industrial uses, Jacek Jakowski, researcher from the University of Tennessee, is accelerating a quantum chemical molecular dynamics method by up to 20x using GPUs, reducing typical simulation times from two weeks to hours.

Researchers Michela Taufer and Sandeep Patel at the University of Delaware are using GPUs to accelerate discoveries in human pathological conditions. With 10x faster simulations, researchers are running much larger size simulations, leading to more accurate insights.

Parallel programming has now become much easier with the recent release of CUDA 4.0 toolkit for C, C++ and Fortran. CUDA 4.0 represents a major leap forward for ease of GPU programmability and offers GPU-accelerated libraries such as FFT, BLAS, LAPACK, RNG, and SPARSE. CUDA C++ support has been enhanced by the powerful Thrust C++ template library, accelerating popular functions such as sorting and reductions.