GPU Profiling for Mobile Application Rendering
Stuttering UI at 30 FPS on a flagship device? We've seen it. Every frame has 16.67 ms to render at 60 FPS. The CPU prepares rendering commands, passes them to the GPU, and the GPU draws. If the GPU misses the next vsync, the frame is dropped — your user sees jank, directly hurting retention and app ratings. Our engineers with 7+ years of production experience identify and fix rendering bottlenecks, ensuring smooth interfaces even on older devices. Contact us for a comprehensive performance audit with a detailed report and recommendations.
Below we cover key tools and techniques we use in our work.
Which Tools to Use for GPU Profiling?
Android GPU Inspector (AGI)
Google Android GPU Inspector is the most powerful tool for mobile GPUs. It works with Adreno (Qualcomm) and Mali (ARM). Requires a device that supports GPU counter profiling — most modern flagships do.
AGI shows:
- GPU Counters — shader unit load, bandwidth, cache hit rate
- Frame Profiler — breakdown of each frame by draw calls, vertex and fragment shader time
- Memory — VRAM usage, texture cache
A critical metric: Fragment ALU utilization above 90% with low Primitive Assembly indicates the problem is in the fragment shader, not geometry. Simplifying the shader or implementing LOD is the right direction.
Xcode Metal Debugger + GPU Frame Capture
For Metal apps on iOS, use GPU Frame Capture in Xcode. It captures a single frame and breaks it down by draw calls. Shows:
- Each
MTLRenderCommandEncoderand its contribution to frame time - Heatmap — which pixels are drawn how many times (overdraw visualization)
- Shader profiler — execution time of specific shader instructions, with source line references
For UIKit/SwiftUI, use the Core Animation Instrument in Instruments. It shows CATransaction, offscreen-rendering passes (yellow layers in Debug → Color Offscreen-Rendered), and layer composition.
Enable Debug → Color Blended Layers in the simulator: red zones are layers with alpha blending. Each red pixel is drawn twice (or more). A screen with 60% red is a real problem for budget devices.
Profiling on Adreno: Snapdragon Profiler
Snapdragon Profiler (Qualcomm) gives the most detailed picture for Adreno GPUs: L1/L2 cache miss rate, texture cache utilization, ALU stalls. We use it when AGI doesn't provide enough detail or when working with Vulkan apps.
The table below compares the tools:
| Tool | Platform | Key Features | When to Use |
|---|---|---|---|
| Android GPU Inspector | Android (Adreno, Mali) | GPU Counters, Frame Profiler, Memory | Initial diagnostics, general analysis |
| Xcode Metal Debugger | iOS | GPU Frame Capture, Heatmap, Shader Profiler | Detailed frame breakdown, shaders |
| Snapdragon Profiler | Android (Adreno) | L1/L2 cache, Texture cache, ALU | Deep Vulkan/Adreno analysis |
Compared to Xcode, Android GPU Inspector provides twice as many GPU counters, including ALU load and cache misses, allowing more accurate diagnosis of shader problems.
How to Fix Overdraw and Offscreen Rendering?
Overdraw. Use Developer Options → GPU Overdraw on Android, Debug → Color Blended Layers on iOS Simulator. Aim to minimize red/pink areas. Remove unnecessary backgrounds in ViewGroup, set opaque = true where transparency is not needed.
Offscreen rendering passes. A CALayer with cornerRadius + masksToBounds on iOS triggers an offscreen render pass — renders the layer to a separate buffer, then composites to screen. In a list with 50 cells, each with cornerRadius, that's 50 extra offscreen passes per frame. Solution: draw rounded corners via UIBezierPath in drawRect or use a background image with transparent corners.
Oversized textures. A 2048×2048 texture for a 44×44 pt icon loads unnecessary data, wasting cache bandwidth. Use MTKTextureLoader with MTKTextureLoaderOptionGenerateMipmaps: true and proper MTKTextureLoaderOptionTextureUsage for mipmapping — standard for 3D and complex UI.
Additional common problems are listed in the table:
| Problem | Symptom | Solution |
|---|---|---|
| Overdraw | Red zones in Color Blended Layers | Set opaque, remove unnecessary backgrounds |
| Expensive shaders | High Fragment ALU with low Vertex | Simplify shader, use LOD |
| Oversized textures | High VRAM usage | Mipmapping, reduce resolution |
GPU Optimization Checklist
1. Check overdraw via GPU Overdraw/Color Blended Layers. 2. Evaluate Fragment ALU utilization in AGI — if above 90%, simplify shaders. 3. Reduce draw calls (batch rendering, texture atlas). 4. Enable mipmapping for textures. 5. Test on target device — FPS should be stable.From Our Practice: 30 FPS on Adreno 650
A card game in Unity on a Samsung S21 (Adreno 650) ran at 30 FPS instead of the expected 60. AGI showed Fragment ALU utilization at 98%, while Vertex Processing was at 12%. The water fragment shader had 4 texture samples + normal mapping + Fresnel calculation. On mobile GPUs, fragment shaders are more expensive than vertex shaders.
Solution: a simplified water shader for the mobile platform (2 texture samples instead of 4), Fresnel approximated via dot(viewDir, normal) without pow(). FPS rose to 58–60 stable. This case demonstrates our approach and experience with Unity and mobile GPU.
How We Conduct Profiling: Step by Step
- Collect metrics on the target device using AGI, Xcode Frame Capture, and Snapdragon Profiler.
- Analyze frame time, identify draw calls and shaders consuming more than 10% of frame time.
- Measure overdraw and offscreen rendering (visual indicators).
- Optimize: simplify shaders, reduce overdraw, cache textures.
- Repeat measurements and produce a final report with recommendations.
What the Work Includes
- GPU performance diagnostics using AGI, Xcode GPU Frame Capture, Snapdragon Profiler.
- Identification of bottlenecks: overdraw, expensive shaders, oversized textures, incorrect tiling.
- Development and implementation of optimizations: shader simplification, overdraw reduction, texture caching.
- Final testing on target device with a guarantee of achieving 60 FPS.
- Report with detailed description of found issues and completed changes.
Timeline and Cost
Profiling and analysis take 2–3 days. Rendering optimization based on results takes 3–10 days depending on complexity. Cost is estimated after reviewing your project — contact us for a free consultation. Order GPU profiling: we guarantee 60 FPS on your device. Cost-per-frame reduction of up to 40% on shader optimization is a real result from our projects.
Our team has over 5 years of experience in mobile development and has completed more than 30 performance optimization projects. We guarantee quality results and transparent interaction at all stages.
Android GPU Inspector documentation — official documentation for in-depth study.







