Profile and optimize AI workloads with Arm Performix
Date: November 4, 2026
Time: 9 a.m. PT | 5 p.m. GMT | 6 p.m. CET
Length: 45 minutes + Q&A
What is limiting your AI inference performance—and how do you turn profiling data into a targeted code improvement?
In this 45-minute code-along we’ll demonstrate an end-to-end optimization workflow using Arm Performix and llama.cpp. You’ll run llama-bench with a 2.5-billion-parameter model, establish a baseline, profile the workload, and use profiling evidence and LLM-assisted analysis to identify a bottleneck and guide an optimization.
You’ll learn:
- How to use Arm Performix and connect it to an LLM for AI-assisted performance analysis.
- How to turn profiling analysis and an optimization hypothesis into a targeted software improvement.
- How to validate changes with repeatable benchmarking and before-and-after comparisons.
What you’ll need:
- Access to an LLM, such as Claude, ChatGPT, or a local model.
- An LLM client that supports connecting to MCP servers.
Host
David Haikney
Technical Product Director, Arm
David is a Technical Product Director at Arm with more than 20 years of experience in performance engineering. His career spans low-level optimization, large-scale infrastructure, and leading teams optimizing mission-critical workloads. At Arm, he is responsible for Arm Performix.
