prudhvi vatala

Snap's head of engineering platforms reports 76% job cost reduction, 62% fewer cores, and 80% memory reduction after migrating 10+ petabytes/day experimentation platform to NVIDIA Spark RAPIDS with zero code changes, calling NVIDIA's direction 'phenomenal' for GPU-accelerated data processing.

Snap's engineering leader calls Google Cloud Dataproc and GKE 'phenomenal' and 'fantastic partner' for scaling to 10+ petabytes/day, praising the three-way partnership with NVIDIA and Google Cloud that enabled 8-9 month production migration of 10+ petabyte/day pipelines.

Snap's head of engineering platforms describes operating at 940M monthly active users with 10+ petabytes/day experimentation platform, achieving 76% cost reduction via GPU-accelerated Spark on Google Cloud with NVIDIA, and building internal platform to share idle inference GPU capacity for batch workloads.

Snap's experimentation platform processing 10+ petabytes daily achieved 76% cost reduction, 62% fewer cores, 80% memory footprint reduction, and eliminated 120TB disk spill by adopting Nvidia Spark RAPIDS on Google Cloud, requiring zero code changes and enabling creative GPU capacity sharing between online inference and batch workloads.

Snap operates at the intersection of AR, AI, and visual communication with nearly 1 billion monthly active users, having pioneered Stories format and Spectacles AR glasses years ahead of competitors, while building a custom GPU-accelerated data platform that enables any internal team to leverage idle inference capacity for batch processing.

Google Cloud's Dataproc and GKE provided the managed Spark runtime and Kubernetes infrastructure that allowed Snap to prototype and productionize Nvidia Spark RAPIDS across 10+ petabytes daily workloads in 8-9 months, with on-demand GPU capacity and seamless fallback mechanisms between GPU, CPU, and Dataproc clusters.