Priyank Patel, Cloudera | Nvidia GTC 2024
summary
Cloudera manages 25 exabytes across hybrid clouds and argues that bringing AI compute to the data beats moving data to models. The stack ties NVIDIA DGX Clouds for pretraining to on‑prem eight‑GPU pods for tuning and inference, uses NVIDIA NIMs as a higher‑level API over GPU primitives, and adopts vector stores for retrieval. Model tuning and deployment require roughly 100–1000x less compute than pretraining, shaping a hybrid operating model.
AI-written summary · about
chapters (10)
- 0:00Introduction to GTC 2024
- 1:50AI and data management evolution
- 3:41Cloudera's hybrid data platform
- 5:50Open-source models vs proprietary LLMs
- 7:40NVIDIA NIMs and developer productivity
- 10:06Data accessibility for AI workflows
- 12:58AI model training and inference split
- 14:23Customer transformation in AI era
- 17:33Generative AI's mainstream impact
- 18:57Future of AI and data integration
related talks
30mDesigning a Hybrid AI Platform on Kubernetes
28mFlexible Cloud, Whatever The Workload
28mScaling agentic AI: From infrastructure to enterprise outcomes
34mScaling AI Like a Pro-PepsiCo’s LLM Deployment Strategy on K...
20mHacking the Inference Pareto Frontier
19mScalable Data and AI Solutions with Cloud-Native Architecture
52mSimplifying the adoption of generative AI for enterprises
05mAI and ML are changing traditional business