DEV Community

Cover image for Mobile Game Power Optimization: GPU vs CPU vs Bandwidth vs Screen vs Network (Real Profiling Data Breakdown)
GameOptim
GameOptim

Posted on

Mobile Game Power Optimization: GPU vs CPU vs Bandwidth vs Screen vs Network (Real Profiling Data Breakdown)

Introduction

As mobile games continue to increase in visual complexity and gameplay intensity, power consumption has become a critical performance constraint. Based on collected profiling data, we analyzed several core factors that influence energy usage in real game scenarios.

Using Perfetto-based metrics and supporting profiling tools, we categorize power consumption into five main areas: CPU, GPU, bandwidth (memory traffic), screen, and network activity.

This article summarizes observed relationships between these factors and real-world power behavior in mobile games.


1. GPU: The Most Direct Driver of Power Consumption

GPU behavior shows a strong and direct correlation with power usage.

In tests conducted on a live game under different graphical quality settings, GPU Clocks and GPU power consumption exhibited a clear positive relationship.

When switching from a low-quality (smooth) preset to a higher-quality preset:

  • GPU Clocks increase significantly
  • GPU Power Rails increase accordingly
  • Overall device power consumption rises noticeably

This confirms that GPU load is one of the most sensitive and direct contributors to power consumption in rendering-heavy scenarios.

In practical terms, higher visual quality almost always translates into higher GPU frequency and higher energy usage.


2. Bandwidth: A Major but Often Underestimated Factor

Bandwidth plays a significant role in both memory and GPU-related power consumption.

In a test scenario involving an empty scene with up to 30 layers of transparent objects, we evaluated the impact of texture compression and mipmap usage.

The results show:

  • Enabling mipmaps reduces bandwidth significantly
  • The impact of mipmaps on power consumption is even more noticeable than texture compression alone

After enabling mipmaps (up to 50 layers test condition):

  • Bandwidth dropped from ~6 GB/s to ~1.5 GB/s
  • Overall power consumption decreased significantly
  • GPU-related metrics such as GPU Clocks also decreased

This indicates that bandwidth is not only a memory-side cost but also indirectly influences GPU workload.

Empirical Observation

Based on data fitting:

Each additional 1 GB/s bandwidth increase corresponds to approximately 60–70 mW additional power consumption.

This aligns in magnitude with industry observations (~100 mW per 1 GB/s).

Practical Recommendation

In 3D rendering pipelines, enabling mipmaps for all textures is strongly recommended to reduce bandwidth pressure and energy consumption.


3. Screen Power Consumption: Brightness, Refresh Rate, and Content Matter

Screen power usage depends on multiple factors:

  • Brightness level
  • Refresh rate
  • Display content

Test results show:

  • Higher brightness increases power consumption
  • Higher refresh rates increase power usage (90 Hz consistently higher than 60 Hz under identical conditions)

An unexpected finding is that screen content itself also affects power consumption significantly:

  • Black frames consume less power
  • High-saturation frames consume more power

In most games, screen power consumption typically ranges between:

150–200 mW

Values significantly above this range may indicate abnormal display behavior or configuration issues.


4. Network: Background Communication Still Matters

Network activity contributes measurable power consumption during both Wi-Fi and cellular data transfers.

Key observations:

  • Network downloads increase both network and CPU power usage
  • Typical network-related power consumption ranges between 100–200 mW

In production game tests, even under low-interaction or idle conditions (anti-AFK scenarios), network power consumption can remain unexpectedly high.

This suggests unnecessary or excessive background communication should be investigated and optimized.


5. CPU: Frequency and Thread Distribution Are Critical

CPU power consumption is strongly influenced by operating frequency.

Using Gears for profiling, comparisons between main-thread execution and Job System execution show:

  • Heavy computation on the main thread keeps big CPU cores at high frequency
  • This leads to significantly increased power consumption
  • Moving workloads into Job System reduces big-core frequency and lowers overall power usage

This behavior is expected because CPU power consumption increases non-linearly with frequency due to the voltage-frequency relationship.

Key Insight

CPU power is not only frequency-dependent but also strongly affected by voltage scaling, making high-frequency operation disproportionately expensive in terms of energy usage.

Optimization Guidance

  • Avoid long-term high-frequency usage on big cores
  • Distribute workloads across threads properly
  • Monitor abnormal CPU time distribution in subthreads

In real-world cases, fixing an abnormal network subthread alone reduced total power consumption by nearly 1W.


6. Profiling Consistency Issue: Release vs Debug Builds

During analysis using SimplePerf, a discrepancy was observed:

  • GOT Online reports showed nearly 10× differences in function contribution
  • SimplePerf results showed similar values

This was traced back to build configuration differences:

  • GOT Online used Release builds
  • SimplePerf used Debug builds (for symbol visibility)

Later verification confirmed:

  • Release builds with IL2CPP still allow function-level visibility in SimplePerf
  • Results become consistent with expectations when using proper Release configuration

This highlights the importance of consistent build settings in performance analysis.


Conclusion

Game power consumption does not have a single root cause or “silver bullet”.

It is the combined result of multiple interacting factors:

  • GPU workload
  • Memory bandwidth
  • CPU frequency behavior
  • Screen configuration
  • Network activity

Only through data-driven analysis can developers accurately identify the dominant sources of power consumption and apply targeted optimizations.

Top comments (0)