Latency Measurements for an Emulation Platform on Autonomous Driving Platform NVIDIA Drive PX2

The use of GPUs in the automotive sector is a possibility to provide a lot of computing power for complex algorithms required for autonomous driving. In order to be able to make statements about the real-time capability of algorithms, it is necessary to be able to make precise statements about the runtime behavior. This work deals with the execution time of instructions on the NVIDIA Drive PX2 evaluation platform designed for an automotive environment. It is to be shown what influence block and warp variables have on execution times and whether these values can be used to estimate the total runtime of algorithms. This provides a deeper insight into the functionality of the modern Pascal architecture.