AWS Podcast #724: Accelerated computing: From fraud detection to AI innovation
Jun 9, 2025
Re Alvarez Parmar, a Container Specialist Solutions Architect at AWS, and Sudhir Kalidindi, a Principal Solution Architect focusing on financial services, share insights into the power of GPU-accelerated computing. They discuss how Rivian leverages Amazon EKS for autonomous vehicle optimization. Sudhir highlights real-time fraud detection, processing over 100 billion events annually. The duo also explores architectural strategies that enhance performance while managing costs for cutting-edge AI workloads, demonstrating the transformative impact of these technologies.
AI Snips
Chapters
Transcript
Episode notes
Maximizing GPU Utilization Matters
- GPUs are expensive and underutilized if usage is only around 30%.
- Maximizing GPU utilization is crucial to avoid wasting thousands of dollars on idle compute.
Design AI Architecture Thoughtfully
- Design your AI ML architecture based on workload type, traffic, and success metrics like speed or accuracy.
- Balance cost with performance by selecting compute, storage, and GPU types suitable for your use case.
Rivian's GPU Workload Orchestration
- Rivian uses Amazon EKS with Argo workflows and a stack called JARC (Jupyter, Argo, Ray on Kubernetes) for multi-GPU autonomous vehicle simulations.
- Their infrastructure team balances job scheduling to fully utilize expensive GPU instances efficiently.