DeepInfra, a purpose-built cloud platform for high-throughput AI inference, announced the results of an independent benchmark confirming the NVIDIA Vera CPU outperforms leading CPUs on agentic workloads. The study found the NVIDIA Vera CPU led by up to 2.2x against competitive CPUs from AMD and Intel, validating its design for the next iteration of AI infrastructure. This performance is critical for platforms like DeepInfra, which processes nearly five trillion tokens per week, with close to 30% driven by agentic systems.

The benchmark utilized DeepInfra’s production AI agent and real captured traffic under identical conditions. Key results indicate that the NVIDIA Vera CPU was the fastest of all four architectures tested in every workload category. It measured up to 2.2x faster orchestration than the x86 baseline, doubling NVIDIA’s published claim of 80% faster agentic CPU performance. Additionally, the CPU sustained up to 1.6x more concurrent agents at the same quality of service compared to competing chips.

Nikola Borisov, co-founder and CEO of DeepInfra, emphasized the infrastructure's tuning for production scale. "We have tuned every layer of it for cost, latency, and throughput at production scale," Borisov said. "NVIDIA is building for where AI workloads are headed, and we believe Vera is exactly the kind of hardware solution the next iteration demands."

Benchmark Results Overview

The testing revealed that on identical CPU partitions, the NVIDIA Vera CPU sustained 256 concurrent agents within strict response-time and error targets, versus 160–192 for the competing chips. Furthermore, with all 256 agents running at full load, the CPU’s leftover cores simultaneously served a 20-billion-parameter open-source model faster than an entire previous-generation CPU socket dedicated solely to that task, without slowing the agents.

Ian Finder, Director of Data Center CPU Products at NVIDIA, noted the importance of production validation. "Bringing Vera to market is an incredible opportunity, but performance is ultimately proven in production," Finder stated. DeepInfra has published a technical analysis detailing the full benchmark methodology and complete results.

NVIDIA announced at SIGGRAPH that its NVIDIA Agent Toolkit now includes NVIDIA Omniverse libraries, a collection of software components designed to equip AI agents with the tools and skills to add physical AI capabilities to existing applications. This integration enables AI agents to prepare 3D content for simulation, addressing the need for robots, factories, and autonomous systems to be designed and tested in virtual environments before real-world deployment.

Omniverse Libraries and Capabilities

The new Omniverse libraries extend the NVIDIA Agent Toolkit into 3D and physical AI workflows. The libraries, including ovrtx, ovphysx, and CAD-to-SimReady skills, are openly available on GitHub. They provide callable tools for sensor simulation, GPU-accelerated physics, and simulation-ready asset validation inside existing applications.

A new blueprint for integrating Omniverse libraries in Blender is also available on GitHub, demonstrating how software makers can add agent-ready simulation capabilities into existing 3D applications.

Industry Adoption and Integration

Software makers SideFX and PTC are integrating Omniverse libraries into their 3D applications. SideFX is using OpenUSD workflows, ovrtx, and ovphysx libraries to explore agent integration into its Houdini procedural 3D content creation workflows. PTC’s Onshape CAD and product data management platform is using OpenUSD and ovrtx to connect cloud-native design workflows with physical simulation.

Startups such as ForgeCAD, Lightwheel, Moonlake AI, and Palatial are also adopting these libraries. Palatial is using CAD-to-SimReady skills to automate asset creation, while Lightwheel is using Omniverse Content Agents to generate SimReady assets from text prompts.

Hardware Support and Availability

Workflows built with Omniverse libraries can run locally on systems ranging from compact RTX-powered units with NVIDIA RTX Spark to NVIDIA GB300-powered systems with NVIDIA DGX Station. RTX Spark systems will be available this fall from ASUS, Dell Technologies, HP, Lenovo, Microsoft Surface, and MSI, with models from Acer and GIGABYTE to follow. DGX Station systems are available to order from ASUS, Dell, GIGABYTE, HP, MSI, Supermicro, and Exxact.