Case Study: TuSimple accelerates AI training with Weka Matrix

TuSimple Selects Weka for Autonomous Fleet Vehicle Machine Learning Training

TuSimple, a leader in autonomous truck technology, needed a storage platform that could keep up with its deep neural network training at massive scale. Its existing approach could not satisfy the performance requirements for on-premises AI training, especially for reading millions of camera image files at high bandwidth. To address this, TuSimple selected Weka and its Weka Matrix™ HPC storage system.

Weka implemented Matrix as a flash-based parallel file system to deliver low-latency, high-bandwidth access for TuSimple’s AI and machine learning workflows. The solution distributes data and metadata across the storage infrastructure to keep GPU clusters fully fed, helping TuSimple accelerate DNN training and scale its data catalog more efficiently in a single namespace. TuSimple also reported better infrastructure cost control thanks to Weka’s hardware-independent design and economical object storage capability.

TuSimple

Xiaoling Han

Senior Director