How to Benchmark Your Gaming PC Properly
Average frame rate hides the problem you are trying to find. One percent lows are the number that matches how a game feels.
The short answer
Average frame rate is a poor measure of how a game feels because it conceals the brief drops that produce perceptible stutter, making one percent and 0.1 percent low figures considerably more informative. Benchmarking usefully requires a repeatable test run under identical conditions rather than comparing different sections of gameplay. Comparing results against your own baseline after a change is more valuable than comparing against other people's hardware.
Verified and last updated August 12, 2026.
- Time needed
- 13 min
- Difficulty
- Intermediate
- Read time
- 3 min
Most benchmarking produces a single average frame rate, which is close to the least useful number available.
A game averaging a hundred and twenty frames per second can feel dreadful. One averaging ninety can feel perfect. The average does not distinguish them.
One percent lows are the number
The average of your slowest one percent of frames. This corresponds to how a game feels far better than the average does, and it is where problems show up.
Step 1: Set up an overlay showing the right numbers
VRAM deserves a caution. Tools report allocated memory, not required memory, and games frequently allocate more than they need when it is available. A high figure does not mean you are short.
Step 2: Build a repeatable test run
This is where most amateur benchmarking falls apart. Comparing two different sections of gameplay compares nothing.
A usable test run needs to be:
- The same route or sequence every time, ideally sixty to ninety seconds
- In a demanding area, not a corridor
- With identical settings, verified rather than assumed
- After a warm-up run, discarded, so caches and clocks have settled
- Repeated at least three times, taking the middle result
That warm-up matters more than people expect. The first run after launching includes shader compilation and cold caches, and it will be unrepresentatively poor.
Built-in benchmarks give perfect repeatability and are less representative of real gameplay. Use them for comparing changes; use real gameplay for understanding your actual experience.
Step 3: Compare against your own baseline
Published review figures come from different drivers, patches, background software, thermal environments and test sections. Comparing against them produces confusion rather than insight.
What is genuinely useful is your own before and after, on the same run, for a specific change. That isolates the variable you changed.
Record your baseline now, while things are working, so you have something to compare against when they are not.
Reading the frame time graph
Frame rate is an average over a second. The frame time graph shows each individual frame, and it makes problems visible that averages hide.
A flat line means consistent pacing, which is what you want.
Regular spikes point at a periodic cause: shader compilation, asset streaming, or a background process on a schedule.
Sawtooth patterns often indicate thermal throttling or power limits.
Occasional large spikes are typically streaming or a background application.
The stuttering guide covers acting on each of those patterns.
Interpreting the results
Our verdict
Judge by 1% lows and the frame time graph, not the average
A hundred and twenty average with forty lows is a stuttering mess. Ninety average with seventy five lows is smooth. The second is the better result despite the worse headline number.
The bottleneck guide covers using utilisation figures to identify which component limits you, and the upgrade guide covers acting on that.
What works
What doesn't
Key takeaways
Frequently asked questions
What are one percent lows?
The average of the slowest one percent of frames in a test run, which represents the dips that produce noticeable stutter. A game averaging a hundred and twenty frames per second with one percent lows of forty will feel worse than one averaging ninety with lows of seventy five. This figure corresponds to perceived smoothness far better than the average does.
Do I need paid benchmarking software?
No. Free monitoring overlays capture frame rate, frame times, one percent lows and component utilisation, which covers everything most people need. Paid tools add convenience and analysis features. The measurement quality is not the differentiator for diagnosing your own system.
Should I use a game's built-in benchmark?
They are excellent for repeatability because the run is identical each time, which is exactly what comparison requires. They are less representative of actual gameplay, since built-in benchmarks are often not the most demanding part of the game. Use them for comparing changes, and use real gameplay for understanding what you actually experience.
Why do my results differ from reviews?
Because test conditions differ in numerous ways: driver version, game patch, background software, thermal environment, memory configuration and the specific test section. Small differences compound. Comparing against your own results before and after a change is far more reliable than comparing against published figures.
About the author
PC Hardware & Performance Editor
Marcus has been building, breaking and repairing gaming PCs since 2011, first as a weekend favour for friends and then for eight years as a bench technician at an independent repair shop in Portland, where roughly a third of the job was diagnosing machines that "ran fine yesterday".
Found something wrong? Games patch and fixes go stale. If a step here no longer works, tell us at hello@spyld.com and we will retest it. Read how we test and our editorial policy.