How to Measure Your PC's Gaming Performance: A Practical Guide

Measuring well is what separates knowing from guessing. Three things are enough: a repeatable, demanding scene, a tool that records the time of each frame plus per-thread CPU usage and GPU usage, and changing one variable at a time. This guide goes through the free tools, the numbers that matter, a test protocol and how to read the results, with the focus on online games, where measuring is harder.

What to measure

Measurement	What it says	What it is for
Frames per second (average)	General performance	Quick comparison
Frame time (ms)	How long each frame takes	Seeing stutter and regularity
1 % and 0.1 % lows	The worst frames	Measuring real smoothness
GPU usage (%)	How hard the card works	Telling CPU-bound from GPU-bound
Per-thread CPU usage (%)	Which threads are saturated	Detecting a single-thread limit
VRAM used (MB)	Card memory in use	Detecting stutter from lack of memory
RAM used (MB)	System memory in use	Same
Temperatures and clocks	Whether something heats and drops	Detecting thermal limits
Ping and packet loss	Network quality	Separating network problems from PC ones

The definitions of average and lows, and the relation between fps and milliseconds, are in frames per second and milliseconds; to convert figures, use the fps and ms calculator.

The tools, all free

• Windows Task Manager. Nothing to install: CPU usage (switch the graph to logical processors to see each thread), GPU, memory and disk. It is the starting point.
• Graphics makers’ overlays. NVIDIA’s (in its app) and AMD’s (in its driver software) show frame rate, frame time, GPU and CPU usage, temperatures and VRAM on screen. Handy for live viewing.
• PresentMon. A free, open-source tool from Intel that measures frame durations for the CPU, GPU and display, and latencies, in DirectX, OpenGL and Vulkan games. It includes a capture app with an overlay and exports frame-by-frame data to CSV.
• CapFrameX. Builds on PresentMon to capture, analyze and compare: it computes the average and percentiles (the equivalent of the 1 % lows) and lets you overlay captures from different configurations.
• Sensor monitors (for example HWiNFO or other tools that read temperatures and clocks) to watch whether the CPU or GPU lowers its clock from heat.
• Network utilities: ping, pathping and tracert in a Windows terminal (ping, latency and servers).

A note of honesty: each tool computes metrics slightly differently, so always compare numbers made with the same tool and the same capture settings.

A test protocol that does not mislead

1. Prepare the environment. Close background programs, update drivers and let the PC warm up a few minutes: the first minute of a game with an empty shader cache does not represent the rest.
2. Choose the scene. The busiest zone you will play (a crowded city, a big base). An empty landscape teaches you nothing.
3. Define a route or a spot. A short route of 60 to 120 seconds, always the same, or a fixed spot with the camera turning the same way.
4. Measure the baseline. Three runs with the current configuration. Note average, 1 % low, GPU usage and the busiest CPU thread.
5. Change one thing. One setting, or one part, or one BIOS setting. Never two at once.
6. Measure again, another three runs, and compare medians.
7. Save the results in a simple sheet so you do not rely on memory.

Run	Change	Average	1 % low	GPU	Max CPU thread	Notes
1 to 3	Baseline
4 to 6	Shadows: High → Medium
7 to 9	Upscaler: Quality

Particulars of an online game

An MMO has no built-in benchmark, and the scene changes: players and fauna are not always in the same place. So:

• Do more runs and trust the median.
• Look for large differences (10-15 % or more); small ones get lost in noise.
• Measure at a similar time of day, with a similar server load.
• Note the ping of each run, in case a network spike hides a settings change.

How to read the results

What you see	What it means	What to do
GPU at 95-100 % and low frame rate	The GPU limits	Lower internal resolution, shadows, lighting
GPU under 80 % and one thread at 100 %	The CPU limits	Lower population and draw distance
Frame rate pinned at one number	A limiter or V-Sync	Check the limits (guide)
Good average but very low 1 % lows	Stutter: memory, storage, VRAM	Check RAM, SSD and textures (guide)
Frame rate drops after a few minutes	Heat or power limit	Check temperatures and cooling
Everything fine on the PC, bad in the game	Network or server	Measure ping and packet loss

On a frame-time graph, a flat line is the ideal; isolated spikes are loading hitches (disk or memory); a regular sawtooth points to sync problems or a badly set limiter; and a line that rises over time, to heat.

To decide the limit, the resolution test is in CPU-bound versus GPU-bound, and the review of causes in what a bottleneck is.

Mistakes that invalidate a measurement

• Measuring with other things open. A browser with video or a download changes the result.
• Comparing captures from different tools. Each defines the lows its own way.
• Measuring the first run. The shader and data caches fill on the first load; discard it.
• Changing zone between runs. A different cloud or a bigger crowd can move several milliseconds by themselves.
• Not noting the configuration. A week later you do not remember what was on. Write down the upscaler, the limiter, the memory profile and the driver version.
• Confusing correlation with cause. If the frame rate rises after a restart, it may have been the restart and not the setting. That is why you repeat.

How we measure

At COLONIZED, a space survival MMO in development with Unreal Engine 5 (closed alpha applications open, first waves planned for winter 2026-2027, not yet playable), we measure in pairs: one change, one measurement before and another after, and sometimes undoing the change to see the cost return. Absolute figures between two launches cannot be compared, because the sky alone moves about 3 ms. The full story, with the development build’s figures, is in our guide to optimizing Unreal Engine 5 games. We have not published system requirements.