NVIDIA AI benchmarks have evolved to measure agentic workloads as enterprises accelerate the adoption of agentic artificial intelligence technologies.
The company released the results of AgentPerf, which is the first benchmark designed for agentic AI workloads. These workloads require executing a series of model calls and using various tools while maintaining context across the entire workflow. Notably, the results showed that the systems achieved 20 times higher energy efficiency per megawatt when running these agents.
In addition, the MLPerf 6.0 Training results demonstrated the requirements for training advanced artificial intelligence models. The platform achieved the fastest training times across all benchmark tests, including pre-training tests for Mixture of Experts models. Furthermore, this release recorded the largest scaling of the Blackwell platform in the history of the MLPerf training benchmarks.
Developments in NVIDIA AI Benchmarks
During the HPE Discover conference, NVIDIA and HPE expanded the capabilities of the HPE AI Factory to support agentic enterprise development. This collaboration integrates the NVIDIA Vera CPU, which is designed specifically to support agentic workloads, into the HPE Private Cloud AI. Consequently, this integration provides stable performance and low latency for tool calling and orchestration within the agent workflow.
Meanwhile, the NVIDIA Agent Toolkit is now available within the HPE Private Cloud AI to provide an operating system for agentic AI. This toolkit enables organizations to monitor behavior, enforce governance, and run multi-agent systems independently. Additionally, the companies expanded NVIDIA Confidential Computing to all HPE AI Factory solutions to protect models and data, which is critical for enterprise cybersecurity.
Hardware Integration and Infrastructure
The partnership also includes the integration of NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs and networking technologies. These hardware components are available through HPE AI Factory at Scale, HPE Sovereign AI, and HPE Private Cloud AI. As a result, these networking technologies available through HPE AI Factory enhance high-performance computers and server systems.
Impact on Enterprise AI Deployment
The introduction of these hardware and software solutions aims to simplify how businesses deploy autonomous systems. By utilizing updated NVIDIA AI benchmarks, organizations can measure the actual performance of their digital agents. This measurement helps in optimizing resource allocation and reducing overall power consumption in data centers.
Future Outlook for Agentic Systems
As enterprises transition from basic models to autonomous agents, the demand for specialized infrastructure will likely increase. The combination of the Vera CPU and Blackwell GPUs provides a foundation for these workloads. Ultimately, the standardized testing provided by NVIDIA AI benchmarks will guide future hardware developments in the industry.
