How to Benchmark a Game Properly: A Repeatable Test for Real Results

Changing a setting and feeling that the game is smoother is not a test. Your eyes miss small differences, and the game itself runs differently each time you play it. A proper benchmark gives you numbers you can compare before and after a change, whether that is a new driver, a lower shadow setting or a RAM upgrade. This guide shows a simple method that needs no special skills, and it explains which numbers to look at.

What a benchmark should do

A good benchmark is repeatable. If you run it twice without changing anything, the results should be close. If they are not, you cannot trust any difference you see after a change. Everything below is about removing the sources of noise: a different scene each time, a hot PC, background apps and changed settings.

The tools you can use

Many games have a built-in benchmark that plays the same scene every time. Use it when it exists, because it is the most repeatable option. For games without one, you can record performance while you play.

• PresentMon is a free tool from Intel. Its project page says it captures high-level performance data for graphics applications on Windows, and that it works with DirectX, OpenGL and Vulkan. It records frame durations and latency values, and a console app can write the data to a CSV file you can open in a spreadsheet. The page also says it needs the right user permissions: without them you can get an "access denied" error when it tries to start a trace.
• NVIDIA FrameView measures frame rates, frame times and power use. NVIDIA's page says it shows the data in an overlay and writes it to a log file, and that it uses PresentMon for the analysis. It can also show chip power, so you can compare performance per watt.

Both pages describe a capture and logging workflow, which is what you want for comparing runs. The overlays are handy for a quick look, but a log file lets you check the numbers afterwards.

Step by step

1. Close what you do not need. Browsers, launchers and recording tools can take CPU time. Our guide on background apps that hurt gaming performance shows how to check.
2. Warm up the PC. Play for ten minutes first, so the PC reaches a normal working temperature. A cold PC can give better results than a hot one. If your temperatures look odd, read what temperatures are normal.
3. Pick one scene and keep it. Choose a spot that stresses the game, such as a busy city, and walk the same route each time. Aim for 60 to 120 seconds. Avoid moments where other players or random events change what is on screen.
4. Write down your settings. Note the resolution, graphics preset, upscaling and driver version, so you can repeat the test later.
5. Run it at least three times. Throw away the first run if it is clearly slower, since shaders may still be compiling, as explained in our shader compilation stutter guide. Then compare the other runs.
6. Change one thing, then test again. If you change three settings at once, you will not know which one made the difference.

Which numbers to read

• Average FPS tells you the general level.
• Frame time is how long each frame takes, in milliseconds. A game can show a high average and still stutter if a few frames take much longer than the rest, so look for spikes in the frame time graph.
• The slowest frames, often called 1 percent lows, describe how bad the stutter is. The PresentMon and FrameView pages do not define this metric, so use it as a general idea: if the slow frames are far below your average, the game feels uneven even when the average is fine.

A difference of a few percent between runs is usually normal variation, so do not trust a small change until you have seen it repeat.

Common mistakes

• Testing a different scene each time.
• Comparing a cold PC with a hot one.
• Trusting a single run.
• Changing the driver and a setting together.
• Reading only the average FPS and ignoring stutter.

Checklist

• The same scene or built-in benchmark, 60 to 120 seconds.
• A warm PC and no extra apps running.
• Three or more runs per setup.
• One change at a time, with settings written down.

When you know the numbers, you can use them to raise FPS in a smart way. Our guide to raising FPS in any PC game and the one on finding out if your CPU or GPU is limiting you are good next steps.