ACADEMIA

GE announced at GTC (NVIDIA GPU Technology Conference - San Jose, California: May 14 -- 17) the high performance GRA112 3U VPX rugged graphics board. Designed for demanding graphical applications as well as GPGPU (general purpose computing on a graphics processing unit) applications such as ISR (intelligence, surveillance, reconnaissance), radar and sonar, it features a 384-core NVIDIA Kepler architecture-based EXK107 GPU. Also provided is 2GBytes of GDDR5 memory which, together with the latest NVIDIA GPU, makes the GRA112 significantly more powerful than its predecessor, the GRA111 -- for which it provides a simple, cost-effective technology insertion upgrade.

Kepler is the latest generation of GPU technology from NVIDIA. Previous GE products, including the powerful multiprocessor NPN240 and IPN250, were based on earlier NVIDIA architectures, and upgrades to these platforms will be announced later in the year.

Availability of the pin-compatible GRA112 also provides a technology insertion upgrade path for users of the MAGIC1 Rugged Display Computer, which currently features the GRA111 processor.

GPGPU has been described as a 'disruptive' technology that has enabled the deployment of sophisticated military/aerospace applications that are significantly higher in performance but that can be developed in a shorter time than is possible with alternative development approaches. It is already widely deployed in a broad range of military/aerospace applications.

The GRA112 is truly rugged -- as opposed to solutions that feature 'after the fact' ruggedization, or that attempt to deploy commercial subsystems in harsh situations -- meaning that it can be deployed with absolute confidence in the most challenging military/aerospace environments that are subject to extremes of shock, vibration, temperature and contaminants. Its small size, weight and power (SWaP) requirement mean that the GRA112 is also an ideal solution for applications that are SWaP-constrained such as unmanned vehicles.

The NVIDIA EXK107 GPU features the company's CUDA parallel computing architecture and delivers significant advances in performance per watt and ease of use over previous generations. The 384-core GPU is capable of 622 GFLOPS, 64 GBytes/second internal memory bandwidth, and 16 GBytes/second external PCI Express bandwidth.

The GRA112 features the latest PCI Express Gen 3 interconnect. GE's GPUDirect Peer-to-Peer technology allows data to stream directly onto the GPU, resulting in a 25x or more decrease in transfer latency from sensors to GPUs. This substantially increases the efficiency of GPGPU applications while simultaneously freeing up CPU resources and memory, allowing GPUs to be applied to time-sensitive applications such as electronic warfare that were previously not viable.

"Program managers are under continuous pressure to reduce development costs and time-to-market while delivering maximum performance in the most demanding applications," said Rod Rice, General Manager, Military/Aerospace Products at GE Intelligent Platforms. "The GRA112 provides a level of performance that, in many applications, is substantially superior to what is possible with traditional programming techniques -- allowing developers to implement more sophisticated, more performance-hungry algorithms. It also enables reduced development timescales: the CUDA architecture, with its extensive commercial acceptance, benefits from a supporting development infrastructure that is both wide and deep."

"NVIDIA CUDA technology is enabling development of the most advanced high performance embedded computing solutions for military applications," said Vineet Gupta, Vice President, Global Automotive and Embedded Solutions at NVIDIA. "Combined with our Kepler GPU architecture, the technology will enable platforms like GE's GRA112 to deliver unprecedented levels of capability."

For radar processing, the introduction of double-precision floating point hardware will minimize the impact of artifacts introduced by single-precision processing. For video processing, the GRA112's higher performance enables superior throughput in applications such as video stitching and sensor fusion in wide area surveillance applications. It can allow the incorporation of data from multiple sensor types to build a complete picture of a large geographical area, delivering increased situational awareness and context.

With a rich set of I/O, the GRA112 is designed to serve many of the most common video applications. Dual, independent channels mean that it is capable of driving RGB analog component video, digital DVI 1.0, and HDMI standards. In addition, the GRA112's video input capability allows integration of sensor data using RS170, NTSC or PAL video formats.

Its development environment includes support for OpenCL, OpenGL, NVIDIA PureVideo Technology (PUHD), NVIDIA PhysX and Microsoft DirectX.