독서리서치레이더투자 프레임워크
로그인 / 회원가입
← 읽기로 돌아가기
뉴스중대2 분 소요DisconfirmAI 편집부

Arm Claims ~2x Rack-Level Performance Advantage Over AMD in Agentic AI Benchmark

A Signal65 report states that Arm's AGI CPU achieves nearly double the modelled rack-level performance of AMD's EPYC 9755 in agentic AI workloads, though results rely on power assumptions and lack final hardware validation.

2026-10-19 08:30 GMT+8

According to a report by TechRadar, an assessment conducted by research firm Signal65 with Arm's cooperation indicates that Arm's AGI CPU achieved an average modelled rack-level performance ratio of 1.99x compared to AMD's EPYC 9755 across seven workloads designed to evaluate agentic artificial intelligence. The comparison relies on estimated performance within a fixed power budget rather than testing on final production hardware.

The assessment highlights that lower individual server power consumption allows more Arm systems to fit within the same electricity allocation. A 24 kW rack can accommodate 21 to 23 Arm AGI CPU servers, compared to only 13 dual-socket AMD EPYC 9755 systems. This configuration provides between 5,712 and 5,888 physical cores for Arm versus 3,328 for AMD, depending on rated power assumptions.

Across workload categories, Arm recorded modelled rack-level advantages averaging 1.7x for tooling, 2.0x for agentic workflows, and 2.2x for reasoning and planning. Separate testing using Sandbench showed Arm processors completing cold starts 45% faster, warm starts 27% faster, and worst-case scenarios under oversubscription 41% faster than AMD's EPYC 9965.

It is important to note that these findings are partly based on power-based modelling, and Arm acknowledged that further testing on final production systems remains outstanding. While the data suggests Arm-based servers could challenge x86 infrastructure in agentic AI deployments constrained by rack power limits, it does not establish universal superiority over general-purpose computing or justify immediate large-scale replacement of existing data center servers.

로그인 / 회원가입
독서리서치레이더투자 프레임워크

출처

  • techradar.com↗
리서치 계속하기 →