Group 1 - Nvidia showcased the Vera Rubin POD platform at GTC, focusing on enhancing competitiveness in cluster computing and inference capabilities for Agent applications [1] - The Vera Rubin POD consists of two types of racks: MGXNVL rack for core GPU computing tasks and MGXETL rack for collaborative processing through direct interconnects [1] - A single Vera Rubin 1152 SuperPOD is composed of 16 Vera Rubin NVL72 racks, 2 Vera CPU racks, 10 Groq 3 LPX racks, 2 BlueField-4 STX storage racks, and 10 Spectrum-6 SPX network racks, highlighting a heterogeneous collaborative system architecture [1] Group 2 - The Groq3 LPX rack accelerates decoding with 256 LPU processors, 128 GB on-chip SRAM, and a bandwidth of 640 TB/s, enhancing the performance of the Vera Rubin NVL72 and LPX combination [2] - Under conditions of 400 TPS per user, the combination of Vera Rubin NVL72 and LPX can achieve up to 35 times the TPS improvement per megawatt compared to NVIDIA GB200 NVL72, making it suitable for low-latency, interactive Agent applications [2] Group 3 - The Vera CPU rack integrates 256 Vera CPUs with a high-density liquid cooling design, supporting over 22,500 concurrent reinforcement learning or agent sandbox environments for testing and validating outputs from Vera Rubin NVL72 and LPX [3]