Arm recently shocked the world with news it was moving its own silicon line into production for the first time in the company’s history. Now, the chip designer is looking to make it easier for developers working on its hardware to develop AI applications with a new toolkit.

Dubbed Performix, the platform includes integrated widgets and pre-configured “recipes” to make it easier to optimize the performance of workloads running on its Neoverse lines.

“It replaces the hodgepodge of open source tools,” Alex Spinelli, Arm’s SVP for AI and developer platforms, explained. “They're good tools, but [Performix] brings a lot of these bits and pieces together in one integrated experience that can help you find code hotspots, give a view of all the instruction mixes that you're using.”

In a conversation with SDxCentral prior to launch, Spinelli said the platform allows developers to go from knowing how their code is running to “take full advantage of the latest micro architecture advances.”

Rustle up an AI feast

At the center of Performix is a model context protocol (MCP) server that allows both human developers and agentic systems to run the platform directly from tools such as GitHub Copilot, Kiro, Gemini, and Codex.

The actual development environment itself – which is free for Neoverse users – leverages recipes that are essentially repeatable and are mapped and tuned for specific application profiles. They allow users to access performance optimization tools, like flame graphs, function tables, and call stacks that are specifically configured to meet user demands.

Arm’s modular approach for Performix also allows users to customize their own recipes, with Spinelli adding: “If you do have tools that you love, great, this is something that can kind of sit alongside and be a companion to those.”

Performix comes with five standard recipes, including Code Hotspot, which lets users directly compare workload execution times, and Memory Access for analyzing how workloads interact with the host device’s memory system to unearth potential latency issues.

There’s also a microarchitecture option that lets users characterize and compare across Arm’s line of hardware, including the new AGI central processing unit (CPU). A full list can be found in the chip firm’s technical blog post, with the chip firm planning to launch additional recipes in future releases.

“We're pretty excited [as] we can continue to add recipes and reuse recipes that our customers create,” Spinelli said. “It's something that we think can really evolve and grow with the industry as well.”

In time for the 'renaissance of the CPU'

The Performix launch is timely as the CPU is becoming increasingly integral to running agentic AI workloads, which require more complex computation than the brute force of a graphics processing unit (GPU).

Developers building agentic systems require CPUs to support more latency-intensive needs, while the sizable memory bandwidth and input-output (I/O) found in modern processor architectures move related agentic data with limited lag.

Referencing Performix’s timely introduction, Spinelli said the platform comes as we shift from the “token economy” to the “agentic economy.”

“We want to empower and enable the developers who are actually building applications. And I think it is the renaissance of the CPU," Spinelli said. "There needs to be a really good partnership and collaboration between the CPU and the GPU, or the CPU and the 'x' accelerator. And we're going to see things moving closer to the edge as well.”