The AVCC (Autonomous Vehicle Computing Consortium) and MLCommons (an open, global engineering consortium) have joined forces to develop the industry’s first Automotive Benchmark. The benchmark consists of specifications and benchmarks that will result in open-source software and certification. Several industry leaders, such as Arm, Bosch, cTuning Foundation, KPIT, NVIDIA, Red Hat, Qualcomm, Samsung Electronics, have already participated in the benchmark.
The benchmark is geared towards making sure that machines used in autonomous vehicles (AV) are able to accurately compare different technologies. After all, efficient vehicles are reliant on a common set of Machine Learning (ML) benchmarks, in order to ensure the safety of roads. To develop the benchmark, AVCC and MLCommons are drawing upon the AVCC AI/ML Benchmark Technical Reports and the MLPerf™ benchmark suites developed by MLCommons. By the end of the year, open-sourced software solutions are expected to be created.
The Executive Director of MLCommons, David Kanter, was excited to use their expertise in machine learning to the automotive industry. He believes that the benchmark will help create standards for ever-advancing intelligent and capable vehicles, while simultaneously creating a wealth of innovation to the industry.
The President of AVCC, Armando Pereira, stated that the benchmark is needed in order for OEMs and automotive suppliers to understand the ultimate performance and system resource requirements of their solutions. With a set of common benchmarks to judge upon, decisions for selecting suppliers and investing in projects can be made accurately and objectively.
At this point, the technical work has just started, and the Automotive Benchmark initiative is actively seeking input and participation from all members of AVCC and MLCommons, in addition to other industry players. Any suppliers and OEMs that are looking to use ML/AI technology for their automotive tasks are invited to participate in the establishment and spread of open-sourced software solutions. The goals of the partnership are to ultimately achieve a standardized set of ML benchmarks that can be used industry-wide.
