Quick answer: Frame rate drops mean the game can't finish a frame in the time budget (about 16.7ms for 60 FPS), so something is taking too long, a bottleneck. The fix is to profile and identify whether you're CPU-bound or GPU-bound and which specific work is consuming the frame, then optimize that. Guessing wastes effort; profiling tells you exactly where the time goes.
Frame rate drops, the game running below its target FPS, are a performance budgeting problem: each frame has a fixed time budget, and a drop means something is exceeding it. The mistake most developers make is optimizing blindly, guessing at the cause and optimizing things that aren't actually the bottleneck. The right approach is to profile first, find the real bottleneck, and optimize that.
Why Frame Rate Drops Happen
To hit a target frame rate, the game must complete all its per-frame work within the frame budget (about 16.7ms for 60 FPS, 33.3ms for 30). A frame rate drop means that work is exceeding the budget, there's a bottleneck. The bottleneck is either on the CPU (game logic, physics, AI, draw-call submission, scripting taking too long) or the GPU (rendering too much, expensive shaders, overdraw, too many or too-large textures). Identifying which side is the bottleneck is the crucial first step, because optimizing the wrong side does nothing.
Drops can be steady (the game is always over budget, a general performance problem) or situational (drops only in certain scenes, during combat, when many things are on screen, a specific load spike). Situational drops point at whatever is different about those moments; steady drops point at baseline cost.
How to Profile and Find the Bottleneck
Use a profiler, do not guess. A profiler shows where the frame's time actually goes: how much on the CPU versus GPU, and within each, which systems or operations consume the most. This immediately tells you whether you're CPU-bound or GPU-bound and what the top costs are. The biggest time consumers are your optimization targets; everything else is noise.
For drops that only happen in the field on players' hardware, capture performance data from real devices. Bugnet's performance and game-health monitoring captures how the game actually performs across real hardware and scenarios, surfacing where frame rate drops occur and on what hardware, so you're not limited to what reproduces on your dev machine. A drop that only happens on certain GPUs, or in a specific scene, shows up in the field data with the context to localize it, complementing local profiling.
How to Fix It
Optimize the identified bottleneck. If CPU-bound: reduce expensive game logic, physics, or AI work per frame; cut draw-call counts (batch, instance); optimize hot code paths; move work off the critical path or spread it across frames. If GPU-bound: reduce rendering cost, fewer/cheaper shaders, less overdraw, smaller or fewer textures, lower-cost effects, level-of-detail for distant objects. For situational drops, target what's specific to those moments (the particle system in combat, the object count in that scene).
Set a performance budget and optimize until the frame fits it on your target hardware, including lower-end machines players actually use (a frame that fits on your high-end dev PC may blow the budget on a typical player's machine). After optimizing, verify the frame rate holds across real hardware via field performance data, and watch frame-time (not just average FPS) to confirm you've removed the drops rather than just raised the average while leaving spikes. Profile-guided optimization of the real bottleneck is far more effective than scattered guessing.
Frame rate drops mean a bottleneck is blowing the frame budget. Profile to find whether you're CPU- or GPU-bound and what's eating the time, don't optimize blind.