I was sitting with two executives from Nvidia in the Truman Lounge of the National Press Club. On the screen in front of me was an almost unrecognizable view of a terrorist speedboat taken from the helmet-mounted camera of a Marine in a helicopter. Even the best-trained observer would be hard-pressed to count the four people in the boat. It was impossible to determine the origin of the craft or identify the weapons on the deck of the boat.
Then with the click of a mouse, the video became a solid, stable, high-definition view in which you could see every detail. Next, was a distant shot covering a vast field across which two people were walking slowly. A movement sensing package was supposed to be highlighting the figures and alerting the person monitoring the view.
But as is the case with surveillance cameras, you had to be watching closely to see the people at all. The movement sensing software was designed to detect the people in the image and alert the operator. But by the time it finally noticed, the people had moved almost fully across the screen. A change in the software and the figures were highlighted almost instantly.
The Nvidia people were showing me the results of improvements in imaging now in use by the military and the intelligence establishment, but rarely seen by the public. What the company has done is take itsGPUs (graphics processing units) and turned them into massively parallel processors. While these processors have been available as specialized workstations and servers, now they’re being mounted in helmet cameras, remotely controlled drones, surveillance equipment and simulators. Effectively, Nvidia has taken thecore of a supercomputer and put it into the field.
The technology behind all of this is built into Nvidia’s Tesla M2090 GPU Computing Modules, which are already being used in a variety of applications. The difference is that now the Tesla modules are being put into helmets and Humvees, drones, manned aircraft and other things the intelligence community really doesn’t want us to know about. The result has been a remarkable transformation in remote imaging and remote sensing.
As the Nvidia folks described it to me, the biggest difference is that the initial processing takes place before the devices’ sensors try to transmit the information back to wherever it’s going.
Another Tool in War on Terror?
So for example, a Predator drone, allegedly belonging to the U.S. Air Force (but really belonging to the CIA or some other murky three-letter agency), records images of activity far below. Because of a variety of factors, these images aren’t really all that clear.
But because the drone can be fitted with a GPU-based computing module, the initial image processing can take place in real time before the image is transmitted to the drone’s controllers and perhaps from there to members of the armed forces. This may have been why Osama bin Laden’s courier, who was being tracked by a drone, could be identified from such a long distance. It was the courier who gave the Navy Seals the indication they needed of bin Laden’s whereabouts.
So was Nvidia one of the reasons we found Osama bin Laden? As you can imagine, neither Nvidia, the CIA nor the military will say. But that courier had to be followed somehow and it needed to be done remotely. There are only so many ways that this can be accomplished.
Of course, this is just one example and perhaps only a theoretical example of what you can do with GPU-based parallel processing. Because the current Nvidia technology can put 512 processing cores on a single chip, the opportunities are substantial. NASA is using this technology for space science applications, and Nvidia has published some weather modeling modules on its Website. These are highly complex mathematical processes and could take a very long time on a traditional computer, even a supercomputer.
One feature I was shown is the ability to create extremely complex flight simulations in a few minutes, versus nearly a month of computer time using Xeon processors. The simulation I saw was of a night landing by a jet fighter on an aircraft carrier in bad weather. This is a problem so complex that many pilots simply can’t handle it, and simulators are a way to save lives.
But as the use of GPU computing modules spreads, so does the type of software that you’ll find supporting it. Right now, most applications are based on Nvidia’s CUDA programming. But that could change. There are rumors that Microsoft is looking at Nvidia’s GPU modules for part of its high performance computing initiative, for example. That could mean, among other things, that you could see a really, really fast version of Windows.
Editor’s Note: This story was corrected to state the correct number of processing cores that can be built into a single chip with current Nvidia technology.