Arm has introduced Performix, a new performance analysis toolkit designed to help developers optimise AI agent workloads across the company’s compute platform.
The launch comes alongside the company’s ongoing shift towards supporting increasingly complex and automated AI systems.
Go deeper with GlobalData
Access deeper industry intelligence
Experience unmatched clarity with a single platform that combines unique data, AI, and human expertise.
Performix is being offered as a free toolkit that integrates with contemporary, automated development workflows.
By providing access to system-wide performance metrics, it allows developers and AI agents to identify operational inefficiencies and optimise software running on Arm-based infrastructure. This includes support for both traditional cloud deployments and newer hardware, such as the Arm AGI CPU.
Arm AI and developer platforms senior vice president Alex Spinelli said: “The shift to agentic AI is introducing new performance challenges. Workloads are becoming more complex, spanning multiple components and layers, while traditional tooling remains designed for monolithic, manually analsed applications.
“At the same time, performance tools must integrate into automated workflows and be usable directly by AI agents, with outputs that are structured, repeatable and actionable.”
Performix is intended to provide continuous, expert analysis and actionable insights on key system metrics including memory bandwidth, latency, cache efficiency and CPU utilisation. This approach is intended to move software performance evaluation from a process relying on manual interpretation of low-level hardware data towards automated, repeatable, and structured outputs.
The toolkit’s core is the Arm MCP Server, which interfaces directly with development environments and tools including GitHub Copilot, Kiro, Gemini and Codex. This integration allows developers and AI assistants to initiate analysis from within their workflow and receive results contextualised alongside source code.
Arm claimed that the Arm MCP Server holds expert performance knowledge and tools to assist with both migration and optimisation tasks. It brings performance evaluation directly into agentic workflows and supports a feedback loop that aims to make analysis a continuous process.
Performix is intended as an extension of Arm’s commitment to supporting optimisation spanning from silicon design through to software stacks.
The toolkit collects runtime performance data directly from the hardware, transforming it into guided insights that enable developers to validate, tune and scale workloads efficiently.
Its extensibility and system-wide visibility are designed to support optimisation across a wide range of workloads, without requiring deep architectural expertise from developers. This guidance uses a recipe-based approach meant to focus developer attention on the most impactful performance areas.
The Performix announcement follows the launch of the Arm AGI CPU last month. The AGI CPU, designed specifically for AI data centre and agentic workloads, supports up to 136 Arm Neoverse V3 cores per processor.
It achieves 6GB/s memory bandwidth per core at sub-100 nanosecond latency, and operates with a 300-watt thermal design power.
Arm AGI CPU’s architecture supports a dedicated core for each program thread, which is intended to provide consistent performance under load while minimising idle cycles and throttling. The CPU can be deployed in both air-cooled chassis supporting up to 8,160 cores per rack and in liquid-cooled environments with over 45,000 cores per rack.
