Graphics Card Benchmarks Explained: FPS, 1% Lows and Frame Times
Some links in my articles are affiliate links. I may earn a commission when you buy through them.
Compare the games, settings and frame-delivery measures that match your own setup. Average FPS describes overall throughput; frame times and low-percentile results help reveal uneven delivery. Keep native, upscaled and generated-frame results in separate comparisons.
Check the game version, scene, resolution, preset, ray-tracing setting and rendering mode. Record the CPU and memory configuration as well. A faster CPU can expose GPU differences; an older CPU can hide them. Results from different reviews are useful context, but they are not one controlled test.
For a demanding single-player game, look for results near your intended visual settings. For esports, use a representative competitive preset and pay attention to the CPU limit. A 4K chart does not answer which complete system is best for low-setting high-refresh play.
Use the performance-limit checks if you already own the PC. For a new purchase, the gaming GPU guide separates budget and performance priorities.
Average FPS and frame time
FPS counts frames delivered per second. Frame time measures how long an individual frame takes, usually in milliseconds. At an evenly paced 60fps each frame takes about 16.7ms; at 120fps it is about 8.3ms. These are calculated timing examples, not measurements of a particular card.
The average can hide a brief pause. A frame-time plot shows whether frames arrive consistently or whether occasional frames take much longer. Check the demanding section rather than just the overall number: traversal, new effects or a crowded scene can expose a problem that a quiet camera pan does not.
About 16.7ms per frame.
About 8.3ms per frame.
About 4.2ms per frame.
What 1% lows can tell you
A 1% low summarises a slow part of the frame distribution, but tools do not all calculate it in the same way. Read the review’s method. Compare values produced by the same tool and procedure rather than treating every published 1% low as interchangeable.
A stronger low result can indicate more consistent delivery in that tested scene. It does not identify the cause by itself. Shader compilation, streaming, CPU work, driver behaviour or memory pressure can all contribute to a hitch. Check repeat runs and the frame-time plot before assigning the issue to VRAM or the processor.
If texture changes improve the problem, the VRAM guide helps interpret the next check. If reducing rendering resolution barely changes the frame rate, investigate the CPU/GPU pairing and any frame cap.
Native, reconstructed and generated results
| Reported result | What it includes | What to check separately |
|---|---|---|
| Native rendering | Game rendered at the output resolution. | Preset, ray tracing and anti-aliasing. |
| Upscaled output | Lower internal resolution reconstructed to the output. | Mode, image quality and render rate. |
| Frame-generation total | Rendered frames plus inserted generated frames. | Base rendered rate, latency and visual artefacts. |
A generated total can look smooth on the screen while the underlying game updates remain slower. Conversely, well-configured reconstruction can be an attractive way to raise useful performance. The rendering-feature guide explains the choice without comparing unlike chart bars.
Judge quality in motion: thin objects, fine foliage, rapid turns and newly revealed areas are useful places to look. A still image misses some of the differences that matter while playing.
Use performance per pound with the real price
Divide the relevant performance result by the actual card price when you want a simple value comparison. Use the same workload and include a required PSU or case change. Do not let an attractive average erase a feature or memory requirement you genuinely need.
For a hypothetical example, if Card A scores 80fps at £320 and Card B scores 100fps at £450 in the same test, their averages are 0.25 and about 0.22fps per pound. A is better on that single value measure; B still delivers a higher frame rate. This example describes the calculation, not real products or live prices.
Choose the smallest shortlist that meets your goal. Specification checks rule out unsuitable models, while the use-case buying guide connects that evidence to the complete purchase.
NVIDIA’s description of memory and low-frame-rate measures · NVIDIA frame-generation measurement guidance