Tuesday 21 September 2010

GPGPU-based digital signal processing applications

GE Intelligent Platforms has announced the latest addition to the AXISLib product family. AXISLib-GPU is a high performance library of optimized math and DSP functions that supports rapid development of the growing number of multiprocessing applications taking advantage of the power of GPGPU (general purpose processing on a graphics processing unit) technology and, specifically, the growing range of rugged, NVIDIA® CUDA™-enabled platforms from GE. AXISLib-GPU complements the existing AXISLib family that comprises AXISLib for Power Architecture™-based systems and the recently-announced AXISLib-x86 for systems based on Intel® processors.

Concurrently, GE Intelligent Platforms is also announcing the 6U OpenVPX CUDA Starter Kit, a complete solution designed to enable the simple acquisition of a complete hardware/software environment for CUDA-based application development.

“AXISLib-GPU provides significant help to our customers in fully exploiting the massive potential of GPGPU technology for demanding digital signal processing applications, offering the highest levels of math and DSP performance,” said Rob McKeel, Vice President, Military & Aerospace Embedded Computing at GE Intelligent Platforms. “Even better, it can help them substantially reduce development time, and thus shorten their time to market. This combination of leading hardware technology and advanced software technology will deliver our customers the sustainable competitive advantage they are looking for.”

AXISLib-GPU supports the development (and, ultimately, speeds the deployment) of high performance DSP and multiprocessing applications on GE’s NVIDIA CUDA-enabled GPGPU platforms such as the IPN250, NPN240 and GRA111. Typical applications include radar, sonar, image processing, signals intelligence and ISR (intelligence, surveillance, reconnaissance). It is a set of signal- and vector processing libraries providing over 500 high performance digital signal processing and vector mathematical functions optimized for NVIDIA’s manycore GPUs and created to help developers maximize system and application performance. AXISLib-GPU features standard APIs for code portability, supporting not only current platforms but also future technology upgrades: this, together with GE’s leadership in long term program support, ensures maximum return on investment and minimum cost of ownership.

AXISLib-GPU can operate in a standalone mode or as an integral software module within the AXIS Advanced Multiprocessor Integrated Software environment which includes an integrated graphical user interface, AXISView, providing task level programming for algorithm development and system visualization. The AXISFlow module provides a high performance interprocessor communication library for high throughput, low latency data movement across multiple fabrics (PCI Express™, Gigabit Ethernet, 10 Gigabit Ethernet, Serial RapidIO® and so on): this permits easy scaling from one to many CPU nodes across a single board, multiple boards in a single chassis and multiple system chassis.

No comments:

Post a Comment