From Research to the Edge

Challenge:

A camera technology company identified a fit for their expertise amidst the booming automotive autonomous and driving aid industry. They had developed a state of the art hardware and deep learning algorithms which took advantage of their unique setup.

The platform of choice for running their product was nVidia’s Jetson AGX. The proof-of-concept software was tested only on workstation PCs. Despite them being more powerful, performance was unsatisfactory and far off their target. They approached us for our expertise in HPC and NVidia GPU acceleration.

Solution:

Software Supreme took on the task and first and foremost put the application through a series of profilers that allowed us to better understand the design and inner workings. We were able to pinpoint a number of bottlenecks that were preventing the application from performing well. We devised a plan that addressed these bottlenecks without changing the architecture devised by the client. Our targeted approach reached and even outperformed the targets set by the client by a factor of 2.

Result:

The application which initially was running at 7fps was not steadily running at 60fps. Our experience in HPC proved that the chosen edge device could not only run the client’s application in real-time, but also utilization was not capped, leaving room for additional functionalities.