Android GPU Profiler

A live-streaming GPU profiler that collects and analyzes GPU timing of Android Vulkan / GLES apps in real time.
Built to inspect per-frame GPU cost and bottlenecks via the Android GPU Layer mechanism, without modifying app source code.

---

Project Overview

Android GPU Profiler is a personal R&D project for streaming GPU timing data of Vulkan / GLES apps running on Android devices to a PC in real time, and
visually analyzing per-frame GPU cost, RenderPass / Scope timings, and GPU load trends.

While RenderDoc and Android GPU Inspector focus on deep analysis of captured single frames,
this project focuses on observing the GPU state of a running app over time.

The goal is to quickly identify, during mobile game development, situations such as:

• Causes of frame spikes at specific moments
• GPU cost tracking for major segments: Opaque / Transparent / UI / Blit / Lighting
• Attach to Unity / Unreal / Native apps without source modifications
• On-device GPU timing collection via Vulkan / GLES layers
• Visualize bottlenecks through frame history and Scope timelines

---

Key Features

Real-time GPU Timing Streaming

Per-frame GPU timing data collected on the Android device is streamed to the PC Host app over TCP.
Instead of a single-frame capture, you can continuously observe frame history while the app runs.

Vulkan / GLES Layer-based Attach

The profiler layer is injected into the target app using Android's GPU Debug Layer mechanism.
No app source modification is needed — any build with android:debuggable="true" can be attached externally.

Host GUI

The PC Host app is built on .NET 8 + Avalonia.
Frame strip, Scope Gantt timeline, GPU Counter panel, and Bottleneck Banner let you spot bottlenecks quickly.

Bottleneck Classification

Based on collected timing data, major frame bottlenecks are roughly classified:

• Fragment-bound
• Vertex-bound
• UI-bound
• Overdraw-heavy
• Blit / Resolve cost
• Lighting cost
• URP pattern estimation

Attach Automation

Selecting a target package in the Host app automates the following:

• Layer APK installation
• Android GPU Debug Layer configuration
• adb forward setup
• Target app launch
• Configuration cleanup on Detach

---

Why I Built It

In mobile game optimization, detailed analysis of a single frame matters, but
quickly understanding "why did that frame just spike?" during actual play is often more important.

RenderDoc and AGI are strong at deep single-frame analysis, but
for watching the full frame flow and seeing when and in which pass bottlenecks recur, they fell short.

This project was built to address that gap.

• I want to see on-device GPU spikes over time.
• I want to attach Unity URP / Unreal / Native apps without engine modifications.
• I want to see per-frame Scope cost in real time.
• I want likely bottleneck sections summarized automatically.
• I want a practical interpretation of mobile GPU pass costs, which are often ambiguous.

---

Tech Stack

Native Layer

• C++17
• Android NDK
• Vulkan Layer
• GLES Layer
• ARM64-v8a
• TCP Server
• GPU Timer Query

Host App

• C#
• .NET 8
• Avalonia UI
• TCP Client
• Real-time timeline rendering
• Frame / Scope data visualization

Android

• Android 9+
• Vulkan 1.1+
• GLES 3.0+
• GL_EXT_disjoint_timer_query
• ADB / Android Platform Tools

---

Structure