Arm AGI CPU Posts 1.99x Modelled Rack Performance in Agentic AI Benchmark

Arm’s AGI CPU delivered an average modelled rack-level performance ratio of 1.99x compared with AMD’s EPYC 9755 across seven agentic AI workloads, according to an assessment by Signal65 conducted with Arm’s cooperation. Individual results ranged from 1.43x to 3.19x.

The workloads included software tools, autonomous task execution, information retrieval and AI model serving, as well as Terminal-Bench 2 and SWE-bench. Signal65 compared measured processor results, then estimated performance across racks operating within a fixed power budget. The findings rely partly on power-based modelling; testing on final production hardware remains outstanding.

For a 24 kW rack, the assessment estimated capacity for 21 to 23 AGI CPU servers, compared with 13 dual-socket EPYC 9755 systems. That would mean 5,712 to 5,888 physical Arm cores versus 3,328 for AMD, based on the report’s rated power assumptions. Arm server power was listed at 950 to 1,050 watts, while AMD’s dual-socket systems were rated at 1,700 watts.

Modelled rack-level advantages averaged 1.7x for tooling, 2.0x for agentic workflows and 2.2x for reasoning and planning. The report also estimated 30% to 35% fewer racks for equivalent database and vector-search workloads.

In separate Sandbench tests against AMD’s EPYC 9965, Arm recorded 45% faster cold starts, 27% faster warm starts and a 41% faster worst-case result under fourfold oversubscription. Arm said further testing on final production systems would include more workloads and comparisons. The results do not establish an advantage across general-purpose computing or complete data center deployments.