Performance Implications of Multi-Chiplet Neural Processing Units on Autonomous Driving Perception
This work addresses performance bottlenecks in autonomous driving perception for automotive systems, representing an incremental improvement in scheduling methods.
The study tackled the challenge of efficiently deploying autonomous driving perception workloads on multi-chiplet Neural Processing Units by proposing a novel scheduling strategy, resulting in an 82% increase in throughput and 2.8x improvement in processing engine utilization compared to monolithic designs.
We study the application of emerging chiplet-based Neural Processing Units to accelerate vehicular AI perception workloads in constrained automotive settings. The motivation stems from how chiplets technology is becoming integral to emerging vehicular architectures, providing a cost-effective trade-off between performance, modularity, and customization; and from perception models being the most computationally demanding workloads in a autonomous driving system. Using the Tesla Autopilot perception pipeline as a case study, we first breakdown its constituent models and profile their performance on different chiplet accelerators. From the insights, we propose a novel scheduling strategy to efficiently deploy perception workloads on multi-chip AI accelerators. Our experiments using a standard DNN performance simulator, MAESTRO, show our approach realizes 82% and 2.8x increase in throughput and processing engines utilization compared to monolithic accelerator designs.