Optimal Synthesis got a 250x speed improvement by turning to Nvidia graphics processing units.
A research firm in Silicon Valley is using general-purpose graphics processing units to revolutionize flight planning for air traffic control, with the goal of making air travel safer by reducing congestion in the skies.
At any given moment, there are around 7,000 aircraft flying in the skies over the United States; that number is expected to double by 2025. Tracking those aircraft with human controllers, even with the help of current traffic control software, is a laborious and occasionally error-prone process.
Optimal Synthesis is working on the ability to analyze air traffic in real-time. The current baseline air traffic control software, developed by NASA, takes about 10 minutes to perform a 24-hour trajectory prediction in four dimensions (latitude, longitude, altitude and time) for 35,000 aircraft (an average day in the air). Optimal Synthesis is using Nvidia’s CUDA programming technology to convert the software for use on a GPU. With the CUDA prototype version, the same application took less than 2.5 seconds, a speedup of 250x.
“GPU computing allows us to exploit the parallelism in the trajectory prediction process,” explains Dr. Monish Tandale. “This in turn allows us to achieve real-time performance and analyze models with greater complexity, and opens up the possibility of utilizing algorithms and approaches that were earlier deemed impractical due to the computational complexity.”
Monish maintains that porting the code to GPUs was relatively easy thanks to the shallow learning curve of CUDA, allowing them to “smoothly transition from C/C++ based CPU programming to GPU programming.” His company is currently looking to exploit GPUs for other government research projects with NASA, the Missile Defense Agency and the Air Force.