skip to content

How do you turn a 60 Hz target into a millisecond frame budget, and why do engineers reason in milliseconds rather than frames per second?

level: juniorimportance: must knowfreq 72%

answer

  1. one second divided by the rate
  2. frame rate is a reciprocal
  3. the same fps drop, different costs
  4. system costs add in milliseconds

basics

~20 s

A 60 Hz target leaves 1000 / 60, about 16.7 ms, per frame for all work. Engineers budget in milliseconds because system costs add up in time, while frames per second is a reciprocal that hides what a change really cost.

solid answer

~50 s

The budget is one second divided by the target rate: `1000 / 60 ≈ 16.7 ms` per frame, `1000 / 30 ≈ 33.3 ms`, and a dedicated server ticking at a fixed 30 Hz gets the same 33.3 ms per tick. A game measured at 45 fps is spending about 22.2 ms per frame, so it is roughly 5.6 ms over a 60 Hz budget, and that 5.6 ms is the size of the problem. Milliseconds are the working unit because costs add: 3 ms of physics plus 2.5 ms of AI is 5.5 ms, whereas frame rates do not add at all. Frames per second is a reciprocal, so the same drop means very different costs: 60 to 50 fps is 3.3 ms, 30 to 20 fps is 16.7 ms. Teams split the budget into per-system millisecond lines and keep some headroom.

go deeper

for a junior

Recall the division, 1000 over the rate, and the anchors: 33.3 ms at 30 Hz, 16.7 ms at 60 Hz, 8.3 ms at 120 Hz.

for a middle

Explain why frame rate is a reciprocal, show that equal fps drops cost different milliseconds, and convert a measured 45 fps into a 5.6 ms overrun.

for a senior

Talk about the budget as a per-frame deadline with headroom, split into per-system and per-pass lines, and apply the same arithmetic to a server's fixed tick.

for a principal

Frame the budget as a contract between teams: who owns each line, how much headroom is held back, and which platform's rate sets the ceiling for shared content.

## What a frame budget is A **frame budget** is the wall-clock time one frame may take if the game is to hit its target rate. Everything the game does each frame has to fit inside it: reading input, running gameplay code, physics, AI, audio, preparing and submitting rendering work, and the GPU drawing the image. The same idea applies to a **dedicated server**. It renders nothing, but it simulates the world at a fixed tick rate, and each tick must finish inside its own **tick budget** or the server falls behind real time. The arithmetic is one division: `budget_ms = 1000 / target_rate_hz` ## The numbers worth knowing | Rate | Time per frame or tick | |---|---| | 30 Hz | 33.3 ms | | 45 fps (a measured rate, not a target) | 22.2 ms | | 60 Hz | 16.7 ms | | 90 Hz | 11.1 ms | | 120 Hz | 8.3 ms | | 144 Hz | 6.9 ms | Two consequences follow at once: - A game measured at **45 fps** spends about 22.2 ms per frame. Against a 60 Hz target it is about **5.6 ms over budget**. That 5.6 ms, not "15 fps", is the size of the problem, and it is the number every optimisation is measured against. - A server ticking at a fixed **30 Hz** has **33.3 ms per tick**. If peak load pushes a tick to 40 ms, the server cannot keep its rate, however comfortable its average tick looks. ## Why milliseconds, not frames per second Frames per second is the reciprocal of frame time, and reciprocals do not behave like costs. Take the same "10 fps drop" at three rates: 1. 120 to 110 fps: 8.33 ms becomes 9.09 ms, about **0.8 ms** more work. 2. 60 to 50 fps: 16.7 ms becomes 20.0 ms, about **3.3 ms** more work. 3. 30 to 20 fps: 33.3 ms becomes 50.0 ms, about **16.7 ms** more work. The headline is identical and the costs differ by a factor of twenty. Costs in milliseconds, on the other hand, add: if physics takes 3 ms and AI takes 2.5 ms on the same thread, together they take 5.5 ms. "Physics costs 20 fps" means nothing, because the frames per second a system appears to cost depend on everything else in the frame. That is why per-system budgets, profiler zones and regression reports are all kept in milliseconds, and why a fps counter is a display for players rather than a tool for engineers. ## Splitting the budget A team turns the total into **budget lines**, milliseconds per system, so each owner knows their ceiling: - the CPU's work is split across systems (gameplay, physics, AI, audio, UI, render preparation), per thread when the engine runs several threads; - the GPU's time is split per pass (shadows, the main geometry pass, lighting, post-processing); - some time is held back as **headroom**, because frame time varies with what is on screen, with background work by the operating system and, on handheld and mobile hardware, with thermal throttling. ## The budget is a deadline, not an average On a fixed-refresh display with vsync on, a frame that misses its deadline does not arrive a little late; it waits for the next refresh. With strict double buffering at 60 Hz, a 17 ms frame misses its refresh and waits for the next one, so the previous image stays on screen for 33.3 ms and the player sees a stutter rather than a 2 % slowdown. How vsync and buffering shape that timing is a game-loop topic; what matters here is that the budget applies to every frame. The server is no different: a tick that overruns delays the ticks queued behind it. ## Common traps - Comparing changes in fps, so a 3 ms win at 60 fps and the same 3 ms win at 30 fps look like different sizes. - Checking the budget against an average frame, when one slow frame in a few hundred is what players notice. - Treating a dedicated server as budget-free because it draws nothing. - Spending the whole 16.7 ms on planned work and leaving no headroom for content that grows after the budget is set.

  • A game runs at 45 fps and must reach 60 Hz. How much must it cut, and does cutting 5.6 ms of CPU work guarantee 60?
    1000 / 45 is about 22.2 ms and 1000 / 60 is about 16.7 ms, so roughly 5.6 ms must go. Cutting CPU time helps only if the CPU is the side that limits the frame; if the GPU also needs 20 ms, the frame stays near 20 ms. The cut has to land on whichever side is over budget, with a margin for variance.
  • Why keep headroom below 16.7 ms instead of planning to fill it?
    Frame time varies with the scene, with operating-system activity and, on handheld hardware, with thermal throttling, and content keeps growing after a budget is set. On a fixed-refresh display a frame just over the deadline is shown late, so a plan that uses every millisecond ships with stutter. How large the margin is remains a team decision.

  • A 10 fps drop costs the same milliseconds at any frame rate.
  • Per-system costs can be budgeted in frames per second.
  • An average of 60 fps means every frame met the 16.7 ms budget.
  • A dedicated server has no time budget because it renders nothing.
  • Cutting CPU time always raises the frame rate of a slow game.