Accelerating Real-Time Analytics with Cosmos DB and GPU-Enhanced Serverless Apache Spark

on YouTube ↗28m

" summary

Shows an end-to-end pipeline that ingests streaming CSV into Azure Storage, trains an XGBoost model with Apache Spark accelerated by NVIDIA Rapids and CUDA, and writes JSON vectors into Cosmos DB where a vector container with 8,124 items is created. Local runs measure 2.38x faster locally GPU versus CPU (typical ~10x when not streaming). Deployment uses Docker images (CUDA runtime), Azure Container Apps (CPU 4 cores 8 GB, GPU 8 cores 56 GB), a shell deploy script, batch writes of 25, and local backups on failure.

AI-written summary · about

" chapters (8)

1. 0:00Introduction
2. 2:32GPU Spark Rapids
3. 7:36CPU versus GPU
4. 10:46Cosmos DB vectors
5. 13:55Azure Container Apps
6. 18:36Docker containerization
7. 21:48Azure CLI deployment
8. 27:54Conclusion and resources for hands-on testing

" related talks

• 30mSpeed Up Your Spark Data Processing with a GPURaphaël Luta · Devoxx France 2021
• 36mVideo analytics at scale: Deep Learning, Computer Vision, ML & SparkClaudiu Barbura · Global Artificial Intelligence Virtual Conference Sep 2020
• 41mAccelerating Data Science with RAPIDSMike Wendt · PyData Los Angeles 2018
• 39mAccelerating Data Science with RAPIDSKeith Kraus · PyData DC 2018
• 19mSpark vs. Accelerated Compute: A Balanced Look at Data Processing StrategiesGiri Tatavarty · Optimized AI Conference 2025
• 30mProcess large-scale data and computational function in serverless mannerReenu Saluja · AzConf 2020
• 43mAnalyzing Big Data in Azure cloud using Spark and DatabricksMaria Zakourdaev · SQLBits 2023
• 46mData Exploration & Experimentation with Notebooks in AzureIan Griffiths · SQLBits 2020