With Groq 3 LPX in Full Production, NVIDIA Extends Vera Rubin Inference for Agents
nvidia ·
NVIDIA announced that its Groq 3 LPX inference accelerator is now in full production as an extension of the Vera Rubin NVL72 platform, achieving 3,400 output tokens per second in a benchmark with an agentic model, reportedly four times faster than competing systems. New industry partners include SpaceXAI, which plans to use NVIDIA Vera CPUs for agentic AI workloads, CoreWeave, which is deploying Spectrum-X Multiplane networking, and Nebius, the first cloud to adopt Groq 3 LPX. The article emphasizes that agentic AI demands low decode latency and large context handling, which NVIDIA addresses by codesigning GPUs and LPUs. It also introduces Spectrum-X Multiplane as a way to scale AI factories to 512,000 GPUs without adding a costly third network tier, while providing fast failure recovery. Additionally, NVIDIA unveiled Scale-In infrastructure powered by BlueField-4 and DOCA to accelerate networking, storage, security, and operations for AI factories.
nvidia