A high FPS counter can sit next to a hitchy camera, a sluggish click or an opponent who snaps across the screen. Natural Selection 2 Build 345 gives a concrete example of how one patch can address all three. Start by checking what went wrong on screen, then measure the part of the game that could explain it.
Start with what feels late
| What you notice | First check | What it helps you isolate |
|---|---|---|
| The whole view hitches, including when you turn the camera | Reproduce it in a local mode and log per-frame times | A stall in producing or displaying frames |
| The view looks smooth, but your click feels late | Repeat a local click and measure when its first visible effect appears | The delay from input to display |
| Your own view responds, but opponents skip or snap back | Keep a frame-time log running and check the match's ping and packet loss around the event | Irregular network updates or server behaviour |
Find the long frame behind a hitch
Average FPS can hide an occasional long gap between frames. In Natural Selection 2 Build 345, the engine began splitting script garbage collection into smaller chunks. Garbage collection reclaims memory that scripts no longer use. When a large cleanup runs in one go on the path to producing a frame, that frame can arrive late even if the surrounding frames arrive quickly.
Use a tool such as PresentMon to log a repeatable route in a local or practice mode. Warm up first and capture long enough for the hitch to recur. Look at the trace for isolated spikes rather than judging the run by the average. Compare the same timing metric across runs. Presented Frame Time tracks the game's frame submissions, while Displayed Frame Time tracks changes on screen where that metric is available.
High FPS but Game Still Stutters? How to Fix Frame Time Spikes
Repeat with a lower render scale while keeping the route, sync settings and FPS cap fixed. Shorter typical frame intervals under lighter graphics load point towards a GPU limit in that scene. If the same hitch survives the change, compare GPU Busy around the spike and check where else the frame could be waiting. A spike alone cannot identify garbage collection as the cause.
How to Identify a CPU Bottleneck in Modern Gaming
Measure local input delay through the display
A steady frame stream can still show the results of older inputs. When the GPU limits throughput, the CPU can queue work ahead of it, so a frame reflecting your latest input may wait behind earlier frames. Reflex Low Latency and Radeon Anti-Lag 2 are game-integrated options that adjust when work starts to reduce that delay on supported hardware. Build 345 added them through its Low Latency option, which the patch shipped with off by default.
In a repeatable local scene, pick an action with an immediate visible cue, such as a muzzle flash. Compare the in-game latency option off and on at the same location, with the same render scale, cap, refresh rate and sync settings. Keep frame generation off if the game offers it. A compatible optical latency tool can capture the click and the first change on the display. For a camera estimate, film the same screen area alongside a marker that lights up when the mouse switch actuates. Count frames from the marker lighting up to the first visible response, then divide by the camera's recording frame rate to get seconds. Repeat each condition and compare the median and spread across clicks.
If you only have a PC Latency reading, record where the metric starts and ends. PresentMon's PC Latency runs from the PC receiving the input to the frame being sent to the display. A physical click-to-screen measurement also includes the input device and when the display visibly changes.
Check how often the server sends new information
A server tick advances the simulation. A server update sends state to clients. Natural Selection 2 Build 345 changed the defaults from 30 to 60 ticks per second and from 20 to 60 outgoing updates per second. At evenly spaced intervals, that change in send rate takes the nominal gap between updates from 50 ms to about 17 ms. Custom servers can use different rates.
Ping measures round-trip network time. Opponent movement on screen also depends on when the server processes and sends an update, how regularly updates arrive, how the client buffers them and when it draws the result. Interpolation can make movement smoother by drawing between received states, at the cost of some delay. A shorter send interval gives the client more frequent chances to receive fresh positions. The interval alone cannot tell you how much sooner an opponent will appear.
If remote players jump while your own camera remains responsive, note when it happens and compare that moment with frame times, in-game ping, packet loss and jitter (variation in network timing) where available. Repeat on the same server if you can. Compare with another server in the same region under similar conditions if the symptom persists. A clean practice run establishes how that local scene behaves. A busy match can add both local load and network variability, so use a trace from the match before deciding which to investigate.
FAQ
Yes. FPS describes how many frames arrive over time. GPU queueing and display timing can change how old the input is by the time a frame is visible. Compare the same local action under both settings with an input-to-screen measurement.
It tells you a frame took longer than usual. Compare the spike with GPU Busy and repeat at a lower render scale. If the hitch persists, investigate CPU-side work or waiting elsewhere in the pipeline before assigning a cause.
These options address local input response in supported games. If opponents snap back while your view still responds smoothly, compare the event with packet loss, jitter and any server performance data the game provides.
The update rate describes how often the server sends state. Your client renders frames on its own schedule. A higher send rate can supply fresher positions even when client FPS stays the same.




