NVIDIA's Vera Rubin Hits Production: 10x Throughput Per Megawatt, 1/10th Token Cost
NVIDIA Blog · rss · 2026-07-21
NVIDIA announced that its next-generation computing architecture, Vera Rubin, has ramped into production, delivering massive leaps in performance and energy efficiency.
Core Performance & Architecture:
- Extreme Efficiency: In a DeepSeek-R1 benchmark run on CoreWeave, Vera Rubin NVL72 achieved 10x more throughput per megawatt compared to Grace Blackwell NVL72, reducing the cost per million tokens to one-tenth.
- Extreme Co-Design: The platform integrates seven chips and five rack trays (including Vera CPU, Spectrum-6 switches) as a single system. The new Vera CPU is optimized for agentic workloads with 2x single-threaded performance.
- Networking & Cooling: The 6th-gen NVLink delivers over 2x throughput. The cable-, fan-, and hose-free tray design cuts assembly time to one minute, and high-temperature dry cooling saves millions of gallons of water annually.
Ecosystem & Partnerships:
- Partners including CoreWeave, Google Cloud, Microsoft Azure, and Oracle Cloud are already deploying the infrastructure.
- Microsoft and Mistral expanded their partnership to build an open-model ecosystem for Sovereign AI in Europe using tens of thousands of Vera Rubin GPUs.
More from Infra
- SmolVM open-sources persistent computer infrastructure for agents that outlive chat sessions — aniketmaurya · 2026-09-11
- PyTorch Day Korea 2026 launches first offline conf, CFP closes Sept 13 — PyTorch · 2026-09-11
- Local LLM server dilemma: 4x CMP-170HX (price up 53% in 20 days) vs Mac Studio M5 Ultra — rumboll · 2026-09-11
- llama.cpp lands Flash Attention tuning for RDNA4, big prefill gains on AMD — pmttyji · 2026-09-11
- Your p99 latency benchmark may be lying: a deep dive into coordinated omission — Franc0Fernand0 · 2026-09-11
- Running MiniMax H3 on 12GB VRAM: quantization, Turbo LoRAs and attention backends compared — Possible_Mood676 · 2026-09-11