Contents
Engine-independent guidance
Don't optimize prematurely — implement simply first
On this page
The principle
Write clear, correct, simple code first; optimize later, and only where measurement demands it. Premature optimization wastes effort on code that will change, obscures the design, and usually targets the wrong thing — "premature optimization is the root of all evil." Correctness and clarity come before speed.
Applies when
Most feature and gameplay code during active development, before the design and content have stabilized enough for the performance profile to be meaningful.
Does not apply / Exceptions
Two important limits. First, architectural decisions that are expensive to reverse — core data layout, threading model, entity architecture — must consider performance up front, because you can't "profile it later" your way out of a fundamentally wrong foundation (this is the counter-tension, and it's real). Second, avoid premature pessimization: deliberately writing gratuitously wasteful code "to optimize later" is its own mistake — don't optimize early, but don't be needlessly slow either. Known hot paths and shipping platform limits also warrant earlier attention.
Rationale
Early in development the game is in flux — systems get rebuilt, content multiplies, the performance profile shifts drastically — so optimizations made before the design settles are often rendered irrelevant or actively harmful by later changes [S-knuth-premature]. Optimizing early also costs twice: once to write the complex fast version, and again to understand and maintain it, all while the simple version would have been fine (only a small fraction of code — Knuth's "critical 3%" — ever meaningfully affects runtime). Clear code is also easier to correctly optimize later, once profiling (PERF-0001) shows where it's actually needed. Simplicity first keeps the code malleable during the iteration that makes the game good (PROTO/ARCH-0005).
Implementation
Default to the simplest correct implementation. Reserve optimization for what the profiler flags (PERF-0001) on the critical path (PERF-0003). Do think about performance for hard-to-change architecture (data layout, PERF-0005; state/serialization, ARCH-0006) early. Otherwise, keep code clear and change it when measurement says to.
Disagreement
The genuine tension: "premature optimization is evil" (defer, measure, keep it simple) vs. "architecture is destiny" (some performance-shaping decisions are effectively permanent and must be made early). The reconciliation most engineers accept: defer local optimizations (algorithms, micro-tuning) until measured, but make structural performance decisions consciously up front. Judge which kind you're facing.
Notes
Pairs with PERF-0001 (measure) and PERF-0003 (bottleneck); its "architecture early" exception connects to ARCH-0006 (serialization/state) and PERF-0005 (data layout). Allies with elegance/simplicity (DESIGN-0007). Confidence 4.
Connected principles
Source trail
Evidence strength describes support for this guidance, not a probability that it is right for your situation.
Citation and record details
GDC-L1-PERF-0002 · contextual guidance · canonical
Recorded verification date: 2026-07-14. This is not a claim of an engine test.