memory-management_skill

Overview

This skill provides practical guidance and reusable patterns for game memory optimization across engines and platforms. It covers object pooling, garbage-collection avoidance techniques, asset streaming, platform memory budgets, and profiling workflows to prevent frame spikes and leaks. Use it to make real-time games more stable and predictable under tight memory and performance constraints.

How this skill works

The skill inspects common hot paths and allocation sources, recommends allocation-free patterns, and supplies concrete implementations such as a generic object pool and an async asset streaming manager. It maps memory layout (stack, heap, static, VRAM) to platform budget guidelines and prescribes a profiling workflow: baseline, monitor, snapshot, trace, fix, verify. It also offers troubleshooting steps for leaks, GC spikes, and out-of-memory issues.

When to use it

• During early architecture design to set platform memory budgets and data layouts
• When frame-time hitches or periodic GC spikes appear in playtesting
• While implementing repeatedly spawned objects (bullets, particles, enemies)
• When streaming large levels or assets for open worlds and VR
• Before mobile and console certification to reduce memory/regression risks

Best practices

• Pre-allocate and pre-warm pools for frequently created objects; cap pool sizes to bound memory
• Avoid allocations in Update/hot paths: use StringBuilder, cached lists, index-based loops
• Profile with engine tools (Unity Memory Profiler, Unreal Insights, native profilers) and compare snapshots
• Design streaming zones with predictive load/unload, async IO, and unload delays to avoid hitches
• Unsubscribe and dispose explicitly; use weak references and IDisposable where appropriate

Example use cases

• Bullet/particle managers using a generic ObjectPool to eliminate per-frame allocations
• Menu-to-game transitions: capture baseline, compare snapshots, and clear scene caches to prevent leaks
• Open-world streaming: load nearby zones async and unload distant zones after a timeout
• Tight mobile targets: follow mobile budget table, reduce texture VRAM, and pre-allocate critical collections
• GC spike troubleshooting: add allocation markers, find hot paths, then apply pooling or struct-based data