Arm’s New Processors Significantly Outperform AMD in AI Workloads
As artificial intelligence continues to evolve, optimizing computing power in data centres has become critical. Recent evaluations indicate that Arm-based processors, specifically engineered for agentic AI tasks, have significantly outperformed AMD’s EPYC 9755 series chips. The study, conducted by Signal65, focuses not only on the capabilities of individual chips but also on the overall efficiency of server racks.
Energy Efficiency as a Key Advantage
According to the research, Arm’s advantage is largely driven by optimized power consumption. Within a 24-kilowatt server rack, it is possible to house between 21 and 23 Arm servers, compared to only 13 dual-socket systems for AMD. This disparity allows Arm to pack significantly more physical cores into a single rack—approximately 5,700 compared to AMD’s 3,328.
Consequently, across seven different types of tests covering agentic AI workloads, Arm systems showed an average performance improvement of 1.99 times. In specific instances, such as data processing and reasoning tasks, this advantage reached as high as 3.19 times. Based on these figures, data centres could utilize 30-35% fewer server racks for similar tasks, significantly reducing operational costs.
Agentic AI and Speed
Modern AI agents constantly create isolated sandbox environments to execute code and retrieve information. Testing demonstrated that Arm processors create these environments 45% faster than competing x86 architecture chips. This speed is crucial for applications where fractions of a second matter.
Arm representatives state that the industry should focus not just on theoretical power, but on how systems handle real-world agentic tasks. Notably, the evaluation framework was developed in collaboration with giants such as Meta, OpenAI, and Fujitsu.
What This Means for the Market
Despite the impressive results, researchers note that these figures are currently based on modelled assessments, and final testing on production hardware is still pending. Furthermore, this Arm advantage is focused on specialized AI workloads and does not imply that the company will automatically replace x86 architecture for general-purpose computing operations.
For technology sectors and users globally, this news is significant as cloud service providers are increasingly shifting toward energy-efficient Arm servers. Ultimately, this will help make AI services cheaper and more accessible, as data centres will consume less electricity while generating more computing power.