NVIDIA collects performance metrics to compare their products to competitive counterparts and develop solutions to any bottlenecks. Power sensing data is crucial to create low overhead programs and minimize resource wastage. One project task was to make power data collection easier by converting NVIDIA's existing power sensing program into a daemon with a well-defined API using named pipes and multi-threaded programming. Another project task was to update NVIDIA's performance scanning infrastructure to support profiling tools. This involved updating its command line to execute a chosen profile, uploading collected data to a database, and developing a method for visualizing that data. We held daily scrums and tested iteratively to implement a power sensing daemon and add profiling support.
Download