Develop

Meta Quest Runtime Optimizer

Updated: Jul 31, 2026

This tool is currently in its experimental phase. Meta is actively gathering feedback from users to shape its future.

The Meta Quest Runtime Optimizer helps developers identify and resolve performance bottlenecks in Meta Quest applications. It provides real-time analysis and actionable insights to optimize GPU and CPU performance, and runs without additional dependencies.

Key capabilities include:

Quick Perf: Real-time CPU and GPU frame time monitoring with automatic bottleneck classification

Bottleneck Analysis: Rule-based analysis of captured frames with per-category insights and render breakdowns

What If? Analysis: Per-GameObject GPU cost measurement via automated A/B experiments

Material and Shader Analysis: Detailed shader instruction counts and cost metrics (requires Adreno Offline Compiler)

Requirements

Meta Horizon OS: V78+

Unity engine version: Unity 2022.3+

Platform OS: Windows 10+ (64-bit)

Unity ID: Create or log in to your Unity account⁠

Installation

Navigate to Meta Quest Runtime Optimizer package⁠ in the Unity Asset Store and click Add to my assets.

In Unity, navigate to Window > Package Management > Package Manager.

Select My Assets and add the Meta Quest Runtime Optimizer package.

Setup and configuration

Step 1: Open the Meta Quest Runtime Optimizer

Navigate to Meta > Tools > Quest Runtime Optimizer in Unity.

Step 2: Enable the Meta Quest Runtime Optimizer

Toggle Quest Runtime Optimizer Enabled on. This updates your project to include all necessary configurations, including enabling Development Build and adding the ENABLE_RUNTIME_OPTIMIZER scripting define symbol.

Disable this option when submitting your application for release candidate (RC) builds.

Step 3: Build your APK

Build your APK in development mode. The tool requires a development build to communicate with the runtime.

Step 4: Verify the APK path

Ensure that your APK’s Executable path looks correct. For example: com.unity.template.vr/com.unity3d.player.UnityPlayerGameActivity

Step 5: Connect the tool

Click Launch to start your application. The tool will automatically connect to your device.

Note: If you start the application a different way, connect to the tool by clicking Connect.

If the application is already running, click Connect to establish the connection.

You should see the “Connected” status message if the tool successfully starts.

Step 6: Freeze the frame to start the analysis

Click Freeze Frame to freeze the current frame. This pauses the application state and takes a snapshot (screenshot and performance metrics) that serves as the baseline for analysis.

Once successfully frozen, the Freeze Frame button updates to Unpause and the Analyze button becomes enabled.

During freeze, the tool disables the proximity sensor and locks head tracking to ensure consistent measurements. There is a 5-second cooldown between freeze operations.

Using the Meta Quest Runtime Optimizer

Quick Perf

Once connected, the Quick Perf view lets you check CPU and GPU frame times. Click Refresh to fetch the latest metrics from the device. The target for VR applications is 14.2ms (~70 FPS). Frame times are color-coded:

Color	Threshold	Meaning
Green	Below 80% of the target	Good performance
Yellow	Between 80–95% of the target	Approaching budget
Red	Above 95% of the target	Over budget

The tool also provides an automatic suggestion based on your frame times:

Scenario	Suggestion
GPU-bound	Use Bottleneck or What If? Analysis to identify costly objects and shaders
CPU-bound	Use Unity Profiler to identify CPU bottlenecks
Both CPU and GPU bound	Reduce overall scene complexity
Good performance	Consider Bottleneck Analysis for further GPU optimization

Use Quick Perf to determine whether your application is CPU-bound or GPU-bound before choosing an analysis mode.

Configuration options

Access the options menu from the toolbar to customize the tool’s behavior:

Option	Description
Enable Adreno Offline Compiler	Enables material and shader cost analysis (requires Qualcomm AOC⁠)
Show Debug Log	Enable verbose debug logging to the Unity Console
Enable PC Testing	Connect to a locally running application on `localhost` instead of via ADB, for testing without a physical Quest headset

The analysis section also has its own options button with a Use Percentage toggle to switch between weight (percentage) and absolute count in vertex and texture analysis views.

Analysis modes

The Runtime Optimizer offers two analysis modes:

Mode	Description
Bottleneck Analysis	High-level direction: identifies performance categories consuming the most GPU budget (~25 seconds)
What If? Analysis	Detailed and precise: measures the exact GPU cost of individual GameObjects (~1 minute, varies with object count)

1. Bottleneck analysis

This mode captures a GPU trace and Unity scene snapshot, then parses the data to highlight potential bottlenecks and provide improvement suggestions.

Ensure your headset remains active during the analysis. If the device enters sleep mode during capture, the tool will detect the empty frame and prompt you to restart the capture.

Use the dropdown menu and select Bottleneck analysis. Then, click Analyze to start the analysis. Once the analysis is complete, click the capture in the Captured Frames section to view the data in the Analysis section.

Insights

The Insights panel provides a summary of captured metrics organized into five performance categories. Each category includes a description of the issue and actionable recommendations:

Category	Trigger	Recommendation
Texture	High anisotropic filtering, texture fetch stalls, or excessive textures per fragment	Combine textures using a Texture Atlas, generate mipmaps, use compressed formats
Fragment	High fragment instruction count per fragment shaded	Use simpler shaders, reduce overdraw and translucent objects
Setup	High draw call count (target: below 300)	Reduce draw call count
Vertex	High vertex instruction count, high MSAA level	Reduce vertex count or shader complexity, lower MSAA level
Vertex Fetch	High vertex fetch stalls or large average bytes per vertex	Reduce vertex data size, vertex count, or draw call count

Render breakdown

The Render Breakdown section organizes runtime data by asset type and displays key metrics for the captured frame:

Draw Call Count: Total draw calls in the frame (below 300 is optimal)

MSAA Level: Current multisample anti-aliasing level

Triangle Count: Total triangles rendered (displayed in thousands)

Loaded Textures Memory: Total texture memory usage (in MB)

Vertex analysis

Lists all meshes sorted by their rendering cost (vertex count * draw count), in descending order. For each mesh, you can see:

Weight or Count: Percentage of total vertex cost or absolute vertex count (toggle via the options menu)

Draw Count: How many times the mesh is drawn per frame

Asset Link: Click to select the mesh asset in the Unity Project window

Texture analysis

Lists all loaded textures sorted by runtime memory size, in descending order. For each texture, the view shows:

Asset Link: Click to select the texture asset in the Unity Project window

Weight or Size: Percentage of total texture memory or size in MB (toggle via the analysis options menu)

Material and shader analysis

Lists detailed material and shader cost data. To access this section, you must enable the Adreno Offline Compiler option and download the Qualcomm Adreno Offline Compiler tool⁠.

When enabled, the material analysis shows per-material shader statistics:

Pixel Shader (PS) instruction count

Vertex Shader (VS) instruction count

Values are color-coded (green/red) to indicate whether they fall within acceptable performance thresholds.

2. What if? analysis

The “What if?” analysis measures the GPU cost of individual GameObjects with high accuracy. It performs automated A/B experiments — disabling each selected object and measuring the before-and-after GPU frame time — similar to manually disabling objects and rebuilding, but without the rebuild step. To start the analysis:

Select What if? analysis from the dropdown menu.

Click Analyze to open the Select Game Objects to Test window.

The window displays all GameObjects currently visible in your camera frustum (not all scene objects). You can:

Search with keywords to filter the list

Select All to select all visible objects (up to 200 maximum)

Use the hierarchical tree view to browse objects by their scene hierarchy

View selected objects in the staging area and remove them individually

Analysis modes

The What If? analysis supports two modes:

Individual mode (default): Each selected GameObject is disabled one at a time. Results show the per-object GPU cost, allowing you to rank objects by their rendering expense.

Group mode: All selected GameObjects are disabled together. Results show the combined GPU cost savings of removing all selected objects at once.

Once you’ve selected the game objects you want to test, click Scan Selected GameObjects to begin the analysis. The results are displayed in the Analysis section, sorted by GPU cost in descending order.

Each result shows:

GPU Frame Time: The GPU time attributable to this GameObject (with ± variance)

GameObject Link: Clickable navigation path to select and highlight the object in the Unity Hierarchy

A negative GPU time is expected when a disabled GameObject was occluding something more costly behind it. This means the object was actually saving GPU time by preventing the rendering of more expensive hidden geometry.

How it works

The test takes approximately 200ms per GameObject. For large selections, the capture timeout automatically extends to accommodate the longer analysis. Results rank all GameObjects by their GPU cost in descending order.

The frame must be frozen first (see Step 6: Freeze the frame to start the analysis). The optimizer locks head tracking to guarantee experiment reproducibility.

Measures the baseline GPU cost of what’s in view.

Iteratively disables selected GameObjects while measuring the before-and-after GPU frame time.

Takes measurements over several frames and averages them for accuracy.

Ranks all GameObjects by their cost in descending order.

Capture management

Captured Frames panel

The left panel displays all captured frames with thumbnail previews and lightweight metrics (FPS, CPU/GPU times, CPU-bound or GPU-bound indicator). Click any capture to view its full analysis.

Export and import captures

You can share captures with teammates or archive them for later analysis:

Export: Click the options button (⋮) on a capture and select Export to save it as a .roz package (a ZIP archive containing the Perfetto trace, screenshots, and analysis JSON)

Import: Click the Import Capture button at the top of the Captured Frames panel to load a previously exported .roz file

Open Perfetto trace

For advanced GPU trace analysis, click the options button (⋮) on a capture and select Open PTrace to open the raw Perfetto trace in the Perfetto UI⁠.

Troubleshooting

Connection issues

Status	Meaning	Solution
Not Connected	No USB device detected	Ensure your Quest is connected via USB and ADB is configured
Connected. Invalid OS	Horizon OS version is below v78	Update your headset to Horizon OS v78 or later
App Running. Not Connected	App is running but TCP connection failed	Click Connect to retry; ensure the APK was built with Runtime Optimizer enabled
Connected. App not launched	Device connected but app isn’t running	Click Launch to start your application

Device enters sleep mode during capture

The tool monitors for device sleep and will show a dialog if detected. The capture is automatically cancelled and you’ll need to restart. To prevent this:

Keep the headset on your head or prop it so the proximity sensor is engaged

The tool automatically disables the proximity sensor during freeze, but this has a 10-minute duration limit

Window loses focus during capture

If the Runtime Optimizer window loses focus during an active capture, you’ll see a warning dialog. Keep the tool window focused and visible throughout the analysis to avoid interruptions.

Empty frame detected

If CPU and GPU frame times drop to near-zero during capture, the tool detects this as a device sleep event and prompts you to restart. Ensure the headset stays awake throughout the analysis.