Spark RAPIDS on GKE

Practical guide and reference materials for running the NVIDIA RAPIDS Accelerator for Apache Spark on Google Kubernetes Engine (GKE).

Based on a production deployment with Apache Spark 3.5.5, RAPIDS Accelerator 25.12.0, and NVIDIA L4 GPUs on g2-standard-32 nodes using the Spark Operator.

Contents

Path	Description
SPARK_RAPIDS_ENABLEMENT.md	Full technical guide
architecture.mmd	Mermaid architecture diagram
examples/	SparkApplication manifests (minimal, production, full)
scripts/	Qualification and log-fetch helpers

Quick start

1. Read the guide: SPARK_RAPIDS_ENABLEMENT.md

2. Pick a manifest under examples/ and adapt image names, buckets, and namespaces

3. Qualify workloads before enabling GPU scheduling:

   ./scripts/qualify.sh gs://<bucket>/spark-events/<app-id>

Reference stack

Component	Version
Apache Spark	3.5.5
RAPIDS Accelerator	25.12.0 (CUDA 12)
GKE node type	g2-standard-32 (NVIDIA L4)
Orchestration	Spark Operator