New research from Intel indicates that its 6th generation Intel Xeon processors deliver significant performance gains for agent-based artificial intelligence systems, with notable improvements seen in healthcare, finance, and manufacturing sectors.
Current trends in artificial intelligence are moving beyond simple query-and-response models towards complex agent systems. These systems are designed to reason, plan, retrieve information, utilize tools, and iteratively refine their outputs. This sophisticated processing generates considerably more system activity than traditional inference workloads, presenting new challenges for IT infrastructure.
“Organizations are increasingly asking a practical question: how much useful work can a system accomplish in a given period? For many agent deployments, throughput becomes the key performance indicator,” noted Intel engineers.
Challenges of Agentic AI
Unlike conventional AI, agent-based systems require efficient management of conversational history, execution context, acquired data, intermediate results, and workflow states. This involves continuous tool usage, access to data sources, launching sub-processes, and coordinating dependent tasks. Consequently, overall performance is dictated not only by computational power but also by the efficiency of data and task movement at each stage of the process.
Intel® Xeon® Advantages for Agentic AI
Intel Xeon processors with built-in accelerators combine robust computing capabilities with substantial memory capacity, high memory bandwidth, and reliable I/O features. This configuration is well-suited to handle the diverse workloads characteristic of agent systems, including high parallelism, significant memory utilization, intensive orchestration, and the continuous execution of varied tasks.
To assess performance, a record-and-replay methodology was employed, ensuring consistent testing conditions regardless of agent dynamics. The evaluation featured a deterministic set of Terminal-Bench tasks, simulating real-world workloads in healthcare, banking and financial services (FSI), and manufacturing, with 24x parallelism over a 60-minute period.
Testing Results
Healthcare: Systems powered by Intel Xeon 6 processors demonstrated superior normalized throughput, outperforming comparable solutions based on 5th Gen AMD EPYC by 1.66x and Arm v9.2A by 4.5x. This advantage applied to both overall throughput and specific task categories: analytics and logic, secure data processing, and control functions.
Banking and FSI: Intel Xeon 6 processors also led in this sector, delivering 1.64x higher throughput than 5th Gen AMD EPYC and 4.57x higher than Arm v9.2A. This enables financial institutions to more effectively manage tasks related to risk analysis, document processing, compliance, and customer service.
Manufacturing: In manufacturing, Intel Xeon 6 processors showed a 1.58x advantage over 5th Gen AMD EPYC and a 4.24x advantage over Arm v9.2A. This contributes to accelerating tasks associated with predictive maintenance, quality control, supply chain optimization, and production planning.
“These results demonstrate that an instance powered by an Intel Xeon 6 processor provides a robust foundation for agentic AI, delivering higher throughput in real-world scenarios across healthcare, manufacturing, and financial services, allowing organizations to process more agent-driven tasks on the same infrastructure,” Intel representatives concluded.
Testing Technical Details:
- Benchmark: Terminal-Bench + Harbor 0.16.1, terminus-2 agent.
- Configurations: Intel Xeon 6 (1-instance m8id.12xlarge: 48 vCPU, 185 GB RAM), 5th Gen AMD EPYC (1-instance m8a.12xlarge: 48 vCPU, 185 GB RAM), Arm v9.2A (1-instance m9gd.12xlarge: 48 vCPU, 185 GB RAM).
- Operating System: Ubuntu 26.04.
- Workload: 24 parallel processes, 60-minute run.
- Primary Metric: Total system throughput (tasks completed per 60 minutes).
Source: Intel
