Nvidia Offers Preconfigured GPU-Based Supercomputing Cluster

Nvidia is rolling out its Tesla GPU Preconfigured Cluster, an off-the-shelf, ready-to-deploy IT infrastructure system that will provide an option for researchers and engineers who are finding it difficult to get time on a supercomputer. The Nvidia Tesla cluster includes a mix of GPU-based systems and x86 CPU servers. Because GPUs use less energy than CPUs, using the Nvidia cluster means both access to computing capacity and reduced power consumption, Nvidia officials say.

Nvidia is looking to make it easier for researchers and IT administrators to add computing capability to their data centers while keeping down energy consumption.

Nvidia and about 20 of its partners May 4 launched the Tesla GPU Preconfigured Cluster, an off-the-shelf system that Nvidia officials said provides up to 30 times more performance than CPU-only solutions.

Additionally, GPU systems consume less power and are smaller and denser than CPU-only systems. The result is that researchers and IT administrators, using the Tesla GPU Preconfigured Cluster, can still offer the compute power needed for HPC (high-performance computing) workloads such as fluid dynamics, seismic processing and financial computing, but at lower costs than CPU-only solutions.

The move is the latest by Nvidia to bring its graphics technology into the supercomputing and HPC spaces, which traditionally are powered by CPUs. According to Nvidia, systems running on GPUs can offer users the computing capabilities they need at a price they can afford.

The Tesla GPU Preconfigured Cluster comes with x86 CPU servers combined with Tesla S1070 1U (1.75-inch) GPU systems. Configurations start at 16 teraflops of performance from four Tesla S1070s, with each system holding four Tesla 10-series GPUs. Each system includes host servers, InfiniBand switches and cabling. They also can be customized to suit the user's needs.

Nvidia officials said there is a growing need for accessible computing offerings in the HPC space, where time on a supercomputer can be difficult to get and buying traditional supercomputers is expensive.

"There are 15 [million] to 20 million engineers, scientists and researchers around the world struggling for time on supercomputers, which has led to a huge pent-up demand for computation," Andy Keane, general manager of Nvidia's Tesla business, said in a statement.

Nvidia's new offering can bring many of these researchers an easy-to-deploy GPU-powered supercomputing cluster that can do the work while reducing power consumption, Keane said.

Nvidia's partners include Appro International, Cray, HPC Technologies, Megaware, Microway and Penguin Computing.