Georgia Tech
researchers are helping the military to unlock the number-crunching
potential of graphics processor cards.
It has long been known that common-or-garden graphics processors
(GPUs), found in every PC, make incredibly fast vector processors -
but they are tough to programme.
"As radar systems and other sensor systems get more complicated,
the computational requirements are becoming a bottleneck," said
Georgia Tech engineer Daniel Campbell. "We are capitalising on the
ability of GPUs to process radar, infrared sensor and video data
faster than a typical computer and at a much lower cost and power
than a computing cluster."
The university is easing programming by writing functions
defined in the Vector, Signal and Image Processing Library (VSIPL)
to run on GPUs.
VSIPL is an open standard developed by embedded signal and image
processing hardware and software vendors, academia, application
developers and government labs.
GPU VSIPL is
available here.
Currently they are writing in Nvidia's CUDA language but,
according to Campbell, the underlying principles can be applied to
GPUs developed by other companies.
With GPU VSIPL, engineers can use high-level functions in C to
perform linear algebra and signal processing operations, and
recompile with GPU VSIPL to take advantage of GPU speed.
The researcher claim that, depending on the function and size of
the data set, VSIPL functions operate between 20 and 350 times
faster on a GPU than a CPU, .
"The results are not surprising because GPUs excel at performing
repetitive arithmetic tasks like those in VSIPL, such as signal
processing functions like Fourier transforms, spectral analysis,
image formation and noise filtering," said research engineer Mark
Richards. "We've just alleviated the need for engineers to
understand the entire GPU architecture by simply providing them
with a library of routines that they frequently use."
US Department of defence has signal and image processing
benchmarks.
"Preliminary studies have shown several of the benchmarks have
straightforward parallelisation schemes that result in faster
operation without requiring significant optimisation," said Georgia
Tech. "For other benchmarks, additional research needs to be
conducted into optimising the use of multiple GPUs."
Plans are afoot to develop additional defence-related GPU
function libraries, and design programming tools for other
processors, such as Cell, found in the PlayStation 3.
Mark Richards (left), Daniel Campbell (centre) and
Andrew Kerr are using PC graphics cards to perform linear algebra
and signal processing.
Mark Richards (left), Andrew Kerr (centre) and Daniel
Campbell have shown that VSIPL functions operate 20-350 times
faster on a PC graphics card than a CPU, depending on the function
and size of the data set.
Georgia Tech researchers have reprogrammed the Vector,
Signal and Image Processing Library (VSIPL) to run on graphics
processing units (GPUs) like this.