← Back
NVIDIA Blog

Up to 30x More Work Per Watt: NVIDIA Vera Rubin NVL72 Sets a New Efficiency Standard for AI Agents

•5 min read•
#compute#inference#nvidia
Level:Advanced
For:AI Engineers
✦TL;DR

NVIDIA’s new Vera Rubin NVL72 GPU delivers up to **30× higher throughput per megawatt** and **35× lower token cost** on agentic AI workloads compared with the prior‑generation GB300 NVL72, as measured with the SemiAnalysis AgentX coding‑session benchmark. The efficiency gains stem from a larger‑scale GPU domain and co‑designed software that together enable up to 40% more GPUs per megawatt via DSX MaxLPS power management. For engineers building production agents, this translates into dramatically more token processing for the same energy budget, making large‑context, multi‑step AI pipelines far cheaper to run at scale.

⚡ Key Takeaways

  • **30× higher throughput per megawatt** on the DeepSeek V4 Pro model versus GB300 NVL72.
  • **35× lower cost per million tokens** compared with GB300 NVL72.
  • NVIDIA DSX MaxLPS technology can provision **up to 40% more GPUs** within the same megawatt budget.
  • Performance measured with the **SemiAnalysis AgentX** workload, which preserves real‑world context growth, tool calls, and sub‑agent spawning.
  • The efficiency advantage applies across multiple agentic models (Kimi K3, MiniMax M3, GLM5.3, Qwen3.5, DeepSeek V4 Pro).
💡 Why It Matters

Engineers can run far larger, long‑context agent pipelines (hundreds of thousands of tokens) without proportional energy cost, enabling cost‑effective deployment of AI agents in power‑constrained data centers.

✅ Practical Steps

  1. Deploy the **Vera Rubin NVL72** GPU in your AI factory to replace GB300 NVL72 for agentic workloads.
  2. Use the **SemiAnalysis AgentX** benchmark suite to validate token‑throughput and power‑efficiency gains on your specific models.
  3. Configure **NVIDIA DSX MaxLPS** power management to maximize GPU density within your megawatt budget.

Want the full story? Read the original article.

Read on NVIDIA Blog ↗

More like this

NVIDIA and Microsoft Expand Accelerated Computing From Azure to Windows PCs

NVIDIA Blog•#compute

The Pulse: What can we learn from the Railway hack?

Pragmatic Engineer•#deployment

Model Compression Beyond Quantization

Machine Learning Mastery•#inference

Into the Omniverse: How Developers Turn Ideas Into Simulations With Frontier AI Agents

NVIDIA Blog•#agents

EXPLORE AI NEWS

Daily hand-picked stories on LLMs, RAG, agents and production AI — curated for engineers who ship.

BROWSE NEWS

GET THE WEEKLY DIGEST

Join engineers getting the Monday signal-over-noise AI breakdown. No spam, unsubscribe anytime.

LEARN AI ENGINEERING

Curated courses, research papers, repos and tutorials built for engineers leveling up in AI.

START LEARNING