Bitclever
aiPublished on June 22, 20264 min read

NVIDIA Blackwell Leads in First Agentic AI Infrastructure Benchmark

The NVIDIA Blackwell Ultra NVL72 platform demonstrates superior performance in AgentPerf, the first benchmark for agentic AI, processing 20x more agents per megawatt.

nvidiablackwellia-agenticabenchmarkagentperfinfraestrutura-aieficiencia-energeticaautomacaollm
NVIDIA Blackwell Leads in First Agentic AI Infrastructure Benchmark
Bitclever AI Research
Author: Bitclever AI Research ## Executive Summary NVIDIA has established a new milestone in agentic AI infrastructure with its Blackwell Ultra NVL72 platform, which demonstrated leading performance in AgentPerf, the industry's first benchmark specifically designed to evaluate agentic AI systems. The results show 20 times higher energy efficiency compared to the previous Hopper generation. ## What Happened Artificial Analysis launched AgentPerf, the industry's first benchmark dedicated to evaluating agentic AI infrastructures. In the first published results, the NVIDIA Blackwell Ultra NVL72 platform stood out as the performance leader across all tested agentic AI workloads. The benchmark reveals a fundamental difference between conversational AI and agentic AI. While completing a chat is comparable to a sprint - one call to the language model (LLM), one response - an agent functions more like a relay race, breaking down an objective into multiple steps and staying active until the task is completed. This complexity results in dozens to hundreds of chained LLM calls, each passing growing context to the next, with tool calls such as code compilation and execution, database searches, and web navigation at each handoff. The complexity is not additive, but multiplicative. ## Why This Matters Traditional AI inference benchmarks only measure a single LLM call: the response speed to a single request and how many simultaneous requests a system can process. These were not designed for agentic workloads, where chained LLM calls, tool call delays, and growing context stress accelerated computing systems in fundamentally different ways. For companies developing and deploying agents at scale, it's crucial to understand agent responsiveness, how many can be deployed simultaneously, and how much useful work the AI infrastructure can deliver for every euro and watt invested. Agentic AI represents a paradigmatic shift in how companies approach automation and decision-making, requiring specialised infrastructures that can support complex and interconnected workflows. ## Business Impact This development has significant implications for organisations planning to implement agentic AI solutions: **Infrastructure Planning**: Companies need to reassess their infrastructure requirements, considering that AI agents require substantially different computational resources from traditional conversational AI models. **Energy Efficiency**: With the 20x increase in energy efficiency demonstrated by the Blackwell platform, organisations can plan larger-scale deployments with more controlled operational costs. **Scalability**: The ability to run multiple agents simultaneously becomes a critical factor for companies looking to automate complex processes across different departments. **ROI and TCO**: Superior performance in agentic tasks allows companies to calculate with greater precision the return on investment in specialised AI infrastructure. ## Bitclever Perspective At Bitclever, we recognise that the transition to agentic AI represents both an opportunity and a challenge for Portuguese organisations. Our expertise in AI, RPA, and business process automation uniquely positions us to help companies navigate this technological evolution. We can assist organisations in evaluating their specific infrastructure needs for agentic AI, developing implementation strategies that maximise the value of technology investments. Our approach integrates performance, energy efficiency, and scalability considerations, ensuring that implemented solutions are aligned with long-term business objectives. Additionally, our expertise in Low-Code platforms such as OutSystems and Appian allows us to create bridges between legacy systems and new agentic AI capabilities, facilitating a gradual and controlled transition. ## Conclusion The launch of the AgentPerf benchmark and the exceptional results of the NVIDIA Blackwell platform mark a decisive moment in the evolution of AI infrastructure. For companies anticipating the implementation of agentic AI solutions, these developments provide clear metrics to evaluate and plan infrastructure investments. As agentic AI becomes mainstream, the ability to measure and optimise performance specific to these workflows will be fundamental to business success in the era of intelligent automation.

Sources

← Back to Blog

Related Articles

Google DeepMind and NVIDIA Revolutionise Local Text Generation with DiffusionGemma
ai3 min read

Google DeepMind and NVIDIA Revolutionise Local Text Generation with DiffusionGemma

Google DeepMind has launched DiffusionGemma, an experimental model that generates text 4x faster than traditional models, optimised by NVIDIA for local GPUs.

Bitclever AI Research
Read article →
ai3 min read

The MANGOS Era: SpaceX, Anthropic and OpenAI Prepare to Revolutionise the IPO Market

The IPO market returns with force, led by MANGOS tech giants. SpaceX, Anthropic and OpenAI prepare public offerings that could redefine market valuations.

Bitclever AI Research
Read article →
UK Advances AI Sovereignty with NVIDIA Technologies: Lessons for Portugal
ai3 min read

UK Advances AI Sovereignty with NVIDIA Technologies: Lessons for Portugal

The UK demonstrates how to transform sovereign AI ambitions into concrete action, doubling AI cloud providers and investing in national infrastructure.

Bitclever AI Research
Read article →