Accelerated computing with GPUs

When the U. S. Government needs help from the military, the armed forces must be fully trained and prepared for anything that might happen. One way to accomplish this is for the military to use lifelike simulations in its training.

The training technologies that have been used in the past have made this process very time consuming, almost to the point of not being worth the time and trouble to use the simulations, experts say.

The technology used in some of these simulations has been vastly improved upon recently, thanks to three companies; San Mateo, CA-based MotionDSP; Newark, DE-based EM Photonics; and Santa Clara, CA-based Nvidia.

This industry used to be very tough to break into. As Sean Varah, CEO of MotionDSP described it, “standards are becoming much more open. The government is making vendors play with each other better. You can’t buy closed systems anymore.”

The U.S. armed forces are now utilizing microchips similar to those being used in PlayStation 3 to generate flight simulations for its aircraft at reduced costs, with more realism, and across a wider array of scenarios. EM Photonics is working with the Naval Air Systems Command on a DARPA-funded project that is testing an algorithm to simulate flight scenarios in a variety of conditions that are too risky to test in real-life. The changes to this technology are undeniable. EM Photonics has cut the compute time of the NASC’s simulation algorithm from 18 hours to 20 minutes, a speed-up of 54x, thanks to the graphic chips from Nvidia which are specially tuned for high performance computing applications.

 EM Photonics uses Nvidia Tesla GPUs to accelerate the simulations necessary to mimic the real-world flight of the Joint Strike Fighter and other aircraft, without putting at risk the lives and machinery of our armed forces.

The same graphic chips are used in unmanned aerial vehicles (UAVs) to process, stabilize and enhance their live streaming videos.

MotionDSP makes the video processing software used by the military for its UAVs. They have improved their technology so that it can now do near real-time processing of any live video source vs. the four to six hours of processing required in its previous systems for the same UAV footage.

MotionDSP uses its Ikena ISR to target Nvidia’s Cuda GPU parallel computing architecture allowing it to render, stabilize and enhance live video faster and better than before. Sumit Gupta, a product line manager at Nvidia, said, “We have a high performance processer, but it’s easy to program. That has been the big success. Cuda is a very easy programming model. We are going to continue to make it easier and easier for more developers to use GPUs.”

The time saved in both the simulation algorithm and the video processing makes the intelligence much more valuable in terms of timeliness and quality. JP Morgenthal, chief architect at Vienna, VA-based Merlin International Inc., the company that sells MotionDSP’s Ikena ISR software directly to DoD and homeland security customers, said that customers are “very excited about what they are able to do in near-real time. This isn’t a capability they’ve had before, so they are still getting adjusted to it. They’re using it as a functional capability right now, but they see the power in it.”

Sean Varah believes, “Our technology makes the impossible possible, and this is making our military safer and better prepared.”