Arm's AI chip nearly doubles AMD's rack performance

Arm's AGI CPU achieves 1.99x average performance over AMD's EPYC 9755 in agentic AI workloads, fitting 23 servers into a 24 kW rack versus 13 AMD systems.

Oct 19, 2026

7 cards · 1m 40s · tap to begin

From · · 1m 40s

Arm's AI chip nearly doubles AMD's rack performance

Arm's AGI CPU achieves 1.99x average performance over AMD's EPYC 9755 in agentic AI workloads, fitting 23 servers into a 24 kW rack versus 13 AMD systems.

In brief

Arm's AGI CPU achieves 1.99x average performance over AMD's EPYC 9755 in agentic AI workloads, fitting 23 servers into a 24 kW rack versus 13 AMD systems. Arm's AGI CPU outperforms AMD's EPYC 9755 by 1.99x on average in agentic AI benchmarks, but real-world adoption depends on software compatibility and deployment…

Arm's AGI CPU beats AMD in AI benchmark

Arm's AGI CPU beats AMD in AI benchmark

Arm's AGI CPU achieved 1.99x average rack-level performance over AMD's EPYC 9755 across seven agentic AI workloads.

Individual workload results range widely

Across the seven tests, Arm's processor results ranged from 1.43x to 3.19x compared with AMD's EPYC 9755.The workloads involved software tools, autonomous task execution, information retrieval, and AI model serving, including Terminal-Bench 2 and SWE-bench.

Rack density gives Arm the edge

A twenty four kilowatt rack could accommodate twenty one to twenty three AGI CPU servers, compared with thirteen dual socket EPYC nine seven five five systems.

More cores fit in the same power budget

More cores fit in the same power budget

That configuration provides between five thousand seven hundred twelve and five thousand eight hundred eighty eight physical cores for Arm, compared with three thousand three hundred twenty eight for AMD.

Arm faster at creating execution environments

Separate testing compared Arm against AMD's EPYC 9965 using Sandbench, which measures time to establish isolated execution environments.

Arm executive on measuring agentic workloads

Arm executive vice president for Cloud AI Mohamed Awad said the industry needs workload specific performance evaluation.

“The industry needs to measure how well systems handle the work agents actually complete”

Real world adoption remains uncertain

The benchmark does not establish that Arm will outperform x86 across general purpose computing or justify replacing existing servers across entire data centers.

The bottom line

Arm's AGI CPU outperforms AMD's EPYC 9755 by 1.99x on average in agentic AI benchmarks, but real-world adoption depends on software compatibility and deployment costs.

Read the original on TechRadar

React

Sign in to react and comment.

Comments (0)

Life is short. Keep it sweet. Respect others' opinions and be kind!

    Recommended next

    More decks on arm and related topics.