Building GPU-accelerated multi-agent apps
summary
Shows a working GPU‑accelerated sustainability intelligence app that pulls satellite imagery, live telemetry and pre‑embedded policy PDFs to produce a decision‑ready risk report; orchestration uses the open‑source Google ADK, vector retrieval with Milvus GPU DB, and serverless inference on Cloud Run with NVIDIA RTX PRO 6000 (G4/Blackwell). Model stack uses Gemma 4 GN-FP4 quantized weights (31B) for multimodal reasoning and Gemma 300M for embeddings; demo runs end‑to‑end and cites Flipkart results of ~50% lower latency and cost when using P2P multi‑GPU.
AI-written summary · about
chapters (16)
- 0:00Countdown
- 1:48Intro
- 3:58What are AI agents enabling?
- 13:40G4 powered by NVIDIA RTX PRO 6000 Blackwell GPUs
- 18:28GPU accelerated sustainability intelligence
- 21:32[Demo] Sustainability Agent
- 27:43What is Model Context Protocol (MCP)?
- 30:03Where is agentic AI being used for workloads?
- 36:56How can developers stop AI agents from getting stuck in a loop?
- 39:39When should organizations transition from human-in-the-loop to agentic AI?
- 42:02Does Gemini Enterprise have access to GPUs?
- 42:50What challenges do AI agents face when retrieving accurate policy data?
- 46:13How do organizations approach security, privacy, and regulations when multi-agent orchestration is used?
- 49:08Where is Jay Rodge using AI agents?
- 53:32How to get started with agentic AI on GPUs
- 56:09Wrap up
related talks
34mGPU as a Service Over K8s: Drive Productivity and Increase Utilization
01h 08mGPU Data Science with Rapids
55mLarge Scale Distributed AI on the JVM
09mDemocratizing Access to GPUs for Faster MLAI Innovation with Generative AI
24mScalable GenAI in Production
23mFrom Gaming to the Metaverse
13mDemocratizing AI with Open Source Compute
25mOpen Source Meets Production MaaS: GMI Inference Engine and Karmada for Global AI…