← Back
NVIDIA
NVIDIA launches Vera Rubin POD, a 60-exaflop AI supercomputer built from five specialized rack systems
· releaseplatformperformance · developer.nvidia.com ↗

NVIDIA Vera Rubin POD Overview

NVIDIA has announced the Vera Rubin POD, a comprehensive AI supercomputer built on the third-generation NVIDIA MGX rack architecture. The platform spans 40 racks with 1.2 quadrillion transistors, nearly 20,000 NVIDIA dies, and 1,152 Rubin GPUs delivering 60 exaflops of compute performance and 10 PB/s of total scale-up bandwidth.

Five Purpose-Built Rack-Scale Systems

The Vera Rubin POD integrates five specialized rack-scale systems, each optimized for different agentic AI workloads:

  • NVL72: Core compute engine with 72 Rubin GPUs and 36 Vera CPUs, optimized for the four scaling laws (pretraining, post-training, test-time scaling, agentic scaling), mixture-of-experts routing, and inference, delivering 4x better training performance and 10x better inference performance per watt vs. Blackwell.
  • Groq 3 LPX: 256 LPUs per rack for extreme low-latency inference
  • Vera CPU: 256 CPUs per rack for large-scale reinforcement learning and sandboxed environments
  • BlueField-4 STX: AI-native storage with CMX for KV cache management
  • Spectrum-6 SPX: Silicon photonics-based networking for low-latency, resilient connectivity

Advanced MGX Architecture Features

The third-generation MGX architecture introduces significant innovations for energy efficiency and reliability:

  • Modular cable-free design for simplified deployment and serviceability
  • Dynamic power steering and intelligent power smoothing for optimized energy distribution
  • Rack-level energy storage for resilience and peak shaving
  • 45C liquid cooling for enhanced thermal efficiency
  • Two copper spine variants: MGX NVL racks with NVLink connectivity and MGX ETL racks supporting either Spectrum-X Ethernet or Groq 3 LPU direct connections

Ecosystem and Deployment

The open MGX standard is supported by more than 80 global partners with established supply chain expertise, enabling fast deployments and seamless scaling. All racks share identical power, cooling, and mechanical envelopes for simplified operations.

Target Use Cases

The Vera Rubin POD is specifically architected for modern agentic AI systems that require planning, tool invocation, code execution, data retrieval, and continuous multistep workflows with numerous AI agents—workloads that generate massive volumes of reasoning tokens and require low-latency, high-throughput coordination across compute, networking, and storage domains.