Runtime Optimizer Part 3 - AI-Assisted Performance Optimization on Meta Quest | Performance Series

Who says performance optimization has to live behind expert-only tooling? In this Start Performance Series session, Meta software engineers Jay Hsia and Nico Lopez show you how the Meta Quest Runtime Optimizer pairs with Perfetto and large language models (LLMs) to turn dense performance data into clear, actionable insights. You’ll also see how Unity developers can use the Runtime Optimizer for GPU analysis, then use Perfetto traces with an LLM to surface bottlenecks and generate specific fixes you can test right away.

💡 By viewing this session, you’ll learn:

• How to use Runtime Optimizer to profile GPU performance in Unity, including deep captures and “what-if” analysis
• How to capture rich Perfetto traces and use them for system-level performance profiling
• How to use an LLM to turn trace data into plain-English bottlenecks and concrete fix ideas
• How to validate improvements with a repeatable before-and-after trace comparison loop

Recorded on December 4, 2025 as part of the Meta Horizon Start program. 

 

🎬 CHAPTERS

👋 INTRODUCTION
🕒 00:00 - What the Runtime Optimizer does and why it matters

⚙️ RUNTIME OPTIMIZER UPDATES
🕒 03:21 - What’s new in Runtime Optimizer 0.2.2
🕒 03:35 - Hierarchy view for GPU cost analysis
🕒 04:14 - Batch profiling game objects

🤖 AI MEETS PERFORMANCE PROFILING
🕒 04:39 - Why Perfetto changes the optimization workflow
🕒 05:32 - Where AI fits into performance analysis
🕒 06:02 - What an MCP adds to LLM reliability

🧪 LIVE DEMO: PERFETTO AND LLM ANALYSIS
🕒 08:42 - Capturing a Perfetto trace
🕒 09:56 - Reading frame breakdowns with an LLM
🕒 10:38 - Detecting anomalies and GPU bottlenecks

🛠 FROM INSIGHT TO FIX
🕒 11:15 - Turning analysis into actionable changes
🕒 12:35 - Applying suggested code changes
🕒 13:02 - Tuning Unity settings with performance context

🔁 THE NORTH STAR OPTIMIZATION LOOP
🕒 16:05 - Runtime Optimizer, Perfetto, and Immersive Debugger together
🕒 17:27 - Running before and after trace comparisons
🕒 18:50 - Measuring real improvements

✅ BEST PRACTICES AND TAKEAWAYS
🕒 19:28 - How to get reliable results from AI-assisted profiling
🕒 20:06 - Why profiling markers matter
🕒 21:00 - Managing tokens and context for better AI results
🕒 22:15 - Applying AI across your profiling toolchain

 

🧰 TOOLS REFERENCED

Perfetto: https://perfetto.dev/

Unity Profiler: https://docs.unity3d.com/Manual/Profiler.html

Cursor: https://cursor.com/

RenderDoc: https://renderdoc.org/

 

📚 RESOURCES

➡️ Fix Performance Bottlenecks with Meta Quest Runtime Optimizer: https://communityforums.atmeta.com/discussions/Community_Resources/fix-performance-bottlenecks-with-the-meta-quest-runtime-optimizer--performance-s/1356116

➡️ Developers Blog: https://developers.meta.com/resources/blog/

➡️ Meta Quest Developer Hub: https://developers.meta.com/horizon/documentation/unity/ts-mqdh/

 

🔗 CONNECT WITH US

➡️ Sign up to get the latest news from Meta Horizon: https://developers.meta.com/horizon/newsletter

 

💡 LEARN ABOUT THE META HORIZON START PROGRAM

The Meta Horizon Start program provides intermediate and advanced developers with the resources, hands-on support, and expert guidance needed to accelerate their app development. Join a thriving community to get the tools and go-to-market guidance you need to successfully deploy and grow your app on Meta Horizon OS. 

Apply to Start today: https://developers.meta.com/horizon/discover/programs/start