Josh Patterson, Voltron Data | Nvidia GTC 2024
summary
Claims a ground-up C++ execution engine that runs everything on NVIDIA GPUs (decompression, CSV parsing, joins, aggregations) and leverages GPU Direct Storage, UCX/InfiniBand and flash to hit multi‑TB and 100+TB queries. Benchmarks state a Theseus distributed query engine runs 10/30/100 TB workloads, plateaus CPU at ~200 nodes/6,400 cores, and that 2 DGX A100s outperform 200-node Spark, delivering roughly a 100X server reduction for large analytic ETL and feature‑engineering pipelines.
AI-written summary · about
chapters (9)
- 0:00The founding of Voltron Data
- 1:43Evolution of data analytics
- 3:24Optimizing GPU performance
- 5:56Scaling for large datasets
- 7:59Use cases and execution
- 10:54Business and licensing model
- 12:59Vision and company strategy
- 14:35Future of digital twins
- 18:31Closing elevator pitch
related talks
20mEnd-to-End Data Science on GPUs
44mArchitecture Effects When Interacting with GPU Technologies
40mFueling the AI Revolution with Gaming
12mAI-powered infrastructure:
55mLarge Scale Distributed AI on the JVM
39mAccelerating Data Science with RAPIDS
23mFrom Gaming to the Metaverse
39mDensity! The not-so-secret trend driving the future of Data Analytics