Newsbite analysis · AI, Data, Hardware · New Product, Software Release
NVIDIA Technical
NVIDIA Vera Rubin and Blackwell Set a New Standard for Agentic AI Performance per Watt
NVIDIA has announced breakthroughs in its Vera Rubin NVL72 GPU, which bolsters agentic AI performance significantly. With the introduction of the AgentX benchmark, the NVL72 shows a robust increase in throughput per megawatt compared to the previous generation. This performance leap is particularly crucial as AI applications demand increasingly efficient processing capabilities. The hardware's design caters to long-context and interactive agent workflows, which are now prevalent in AI workloads.
Act nowConfidence: HighImmediate horizon
Published 5 Sept 2026, 12:40 UTC8 min read
Lead image — NVIDIA Technical
Brief
NewsBite reading:NVIDIA Sets New Performance Benchmark with Vera Rubin for Agentic AI
TL;DR
- Vera Rubin NVL72 delivers up to 30x higher throughput per megawatt than GB300 NVL72.
- AgentX benchmark measures performance for long-context, interactive AI tasks.
- Supports a substantial increase in agentic inference capacity without compromising user experience.
- Optimized for dynamic workloads with efficient resource usage.
- New GPU architecture enhances economics for operators with lower costs per million tokens.
Interesting facts
- AgentX addresses gaps in benchmarking for realistic AI scenarios.
- Vera Rubin's design reflects the shifting landscape of AI workload demands.
What changed
The introduction of the Vera Rubin NVL72 GPU marks a significant shift in performance benchmarks for agentic AI processing, enhancing throughput energy efficiency.