Product Launchlpuvera cpudynamo
Nvidia Prioritizes LPU Decode Over CPX
8.3
At GTC 2026, Nvidia VP Ian Buck said the company is delaying CPX to focus on LPU-based decode this year, integrating Groq 3 LPUs with Vera Rubin and Dynamo software. He showed a Vera CPU reference module — an 88-core dual-socket LPDDR5 design — positioning it for agentic AI training and deployment tasks and noting NVLink Fusion interconnect support.
Key Points
- 1Delays CPX to prioritize LPU-based decode using Groq 3 LPUs and Dynamo software
- 2Explains split decode: LPUs handle SRAM-friendly layers while GPUs perform attention and KV computations
- 3Introduces Vera CPU reference module, 88-core dual-socket design boosting single-threaded agentic AI performance
Scoring Rationale
Official, detailed product roadmap and technical clarity drove score; limitation is narrower focus on data-center and agentic workloads, not consumer market.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems