eNEWSLETTER SIGNUP
Sign up for our free
Industry eNewsletters
Q+A CENTER
Latest question:
How do you think the new GigE standards will influence the machine vision industry?
Respond or ask your question now!
Industry News
Most Read
Most E-mailed
E-mail Article
Print Article
DigitalGlobe Sees the World with NVIDIA CUDA
Satellite imagery is ubiquitous,
providing a rich and intuitive context to the company's web-based, desktop, and
mobile environments in map, real estate, personal navigation, and other
applications,including Google Maps and Google Earth. This data has profound
impact on our lives -- from monitoring agricultural crops and
managing the risks of wildfires to planning hurricane evacuation routes and
rescue efforts after natural disasters.
DigitalGlobe provides accurate,
high-resolution custom imagery commercially available, in a wide variety of
image formats. With gigabytes of data constantly streamed to earth from their
satellites, DigitalGlobe uses sophisticated image re-sampling and interpolation
algorithms to provide customers with custom imagery analysis. This data- and
compute-intensive process traditionally required multiple passes through
DigitalGlobe's CPU-based cluster, slowing customer deliveries.
DigitalGlobe re-architected its image
processing software for the NVIDIA CUDA parallel processing architecture so that all pixel
manipulation now can be done on a power-efficient NVIDIA Tesla S1070 GPU cluster. This
enables mass customization of data for much faster image processing. The
Tesla GPU-based cluster will deliver more than 110 times the compute power of
the CPU-based cluster, literally saving weeks of processing time while consuming
much lower power.
"By using just one Tesla S1070 GPU 1U system we were able to complete nearly a month's worth of image processing for a product line in a single night," says Stuart Schwartz, Product Manager at DigitalGlobe. "With the compute capabilities of the GPU, we are experiencing a much higher performance result at a lower cost and power requirement. It also opens up opportunities for us to experiment with new image processing algorithms that were too expensive to run on a CPU-based cluster."
