Case Study: Leading Automotive Company improves GPU utilization with WekaIO Matrix

Machine Learning for Automotive Improve Utilization of GPU Resources

Leading Automotive Company, a leading autonomous vehicle manufacturer, needed to improve utilization of its GPU platform so it could run more training epochs faster, accelerate time to market, and get better ROI from expensive GPUs. Its existing NFS-based all-flash NAS could not scale to production-sized workloads, struggled with millions of small files and large data sets, and could not keep GPU servers fully supplied with data.

Weka addressed the challenge with WekaIO Matrix™ software on commodity x86 server infrastructure, providing a high-performance, scalable, single-namespace data lake for training data. The results included 7X better performance than NFS-based all-flash NAS, 3X better metadata performance, 50% lower storage tier cost, improved data scientist productivity by removing local NVMe copy steps, and a simpler architecture with integrated tiering and full-bandwidth S3-to-GPU delivery.