How to Read GPU Benchmarks: FPS, Percentiles, and Scores

What benchmarks actually measure

A GPU benchmark is a controlled test that runs a game or a synthetic workload and records performance. The simplest output is average frames per second (FPS), which tells you how many frames the card draws on average over the test. That number feels useful, but it hides frame-to-frame variability.

The more informative numbers are percentiles. The 1% low is the average FPS of the slowest 1% of frames, and the 0.1% low is the average of the slowest 0.1%. These reveal stutter and frame pacing issues that average FPS misses. A card might average 120 FPS but drop to 60 for a few frames, and that dip is what you feel as a hitch.

• Average FPS: total frames divided by test time. Good for a quick comparison.
• 1% low FPS: average of the worst 1% frames. Lower means more stutter.
• 0.1% low FPS: even rarer, heavier hitches. Important for competitive play.
• Frame time graph: shows the time each frame takes; flat lines are smooth.

Synthetic scores versus game benchmarks

Synthetic benchmarks produce a single score, like a 3DMark number. They stress the GPU in a controlled, repeatable way and are useful for tracking overclocks or comparing cards on the same system. However, a synthetic score does not tell you how a card behaves in a specific game, because games use different engines, shading, and memory patterns.

Game benchmarks are closer to what you will experience. They run a real game with a fixed scene and report FPS. The best benchmarks use multiple scenes to cover different types of loads. When you see a game benchmark, check which resolution and quality preset were used. The same card can perform very differently at 1080p versus 4K, and with ray tracing on or off.

Because every benchmark scene is different, you should only compare numbers from the exact same game, scene, resolution, and settings. A chart that lists relative performance, like NVIDIA's comparison page for the RTX 50 Series, does this by using consistent conditions and showing each card's performance relative to a reference. That is a fairer way to see generation-to-generation gains than comparing independent results.

Why test settings change the story

Resolution is the first filter. At 1080p, many cards are CPU-bound, so the difference between a mid-range and a high-end GPU may be small. At 4K, the GPU does more work, and the gap widens. NVIDIA's comparison page lists relative performance at 4K, 1440p, and 1080p separately, which is exactly how you should look at it: pick the resolution that matches your monitor. See our resolution and refresh rate guide for guidance.

Quality presets also matter. "Max settings" can include ray tracing, which heavily loads the GPU. The RTX 50 Series page notes that its relative performance numbers use "Max Settings" with DLSS enabled for some features. That means the benchmark is showing what the card can do with upscaling and frame generation, not pure raster performance. Always check whether the benchmark used DLSS, XeSS, FSR, or frame generation, because those technologies boost FPS at the cost of some image quality.

A card that wins at 1080p without upscaling may lose at 4K with DLSS enabled. The reverse is also true: a card with strong tensor cores may leap ahead when DLSS is on. So you have to read benchmarks in the same conditions you plan to use.

Upscaling and frame generation change the numbers

Upscaling technologies like NVIDIA DLSS and Intel XeSS render at a lower internal resolution and then reconstruct the image. This raises FPS substantially. Frame generation goes further: it creates synthetic frames between real ones, multiplying the frame rate. NVIDIA's DLSS page explains that Multi Frame Generation can boost FPS by generating up to five frames per rendered frame on RTX 50 Series GPUs.

These features make raw FPS numbers misleading if you compare a card running native rendering with a card running DLSS. A benchmark that uses DLSS is measuring the card's AI-assisted performance, not its raw power. When you see a big FPS number, ask which upscaling mode was used and at what quality setting. Performance mode gives the largest boost but the largest quality trade-off; quality mode is more conservative.

For a fair comparison, look for benchmarks that list both native and upscaled results separately. If you plan to use DLSS yourself, then the upscaled number is more relevant. If you never enable it, native numbers matter more. The same logic applies to frame generation: it improves smoothness but adds latency, and not everyone wants that.

How to compare cards fairly

To compare GPUs from benchmark results, you need to control for variables. Use a trusted source that runs cards on the same test bench with the same processor, memory, and drivers. Look for charts that show multiple cards across a set of games, not just a single game. The NVIDIA comparison page does this by showing each card's relative performance against a baseline under a fixed game scenario.

Check the driver version and the date of the benchmark. Graphics drivers change performance, and new releases may add optimizations. Also check the card model: the same chip can vary by factory overclock and cooling, so a custom board may beat a reference card. If you are comparing two cards, make sure they have the same memory size and bandwidth, because lower memory limits can cap FPS in certain scenes.

Finally, consider your own workload. A card that leads in 4K with ray tracing may be overkill for 1080p esports. If you play at 1440p with high refresh, look at 1440p average FPS and 1% lows. If you do content creation, synthetic scores for compute may matter more than game FPS. The right benchmark is the one that matches your monitor and your favorite games.

Apply the results to your purchase

Once you understand the numbers, you can use them to pick a card. Start with your target resolution and desired FPS. For 1080p, even a modest GPU often suffices. For 4K, you need a high-end card with enough memory and bandwidth. You can also read the guides on VRAM size and memory bandwidth to match memory to your needs.

Benchmarks can also help you decide between a card with more shader cores and one with faster upscaling. If a game you play supports DLSS or XeSS, a card that excels at AI-assisted rendering might give you double the FPS. That is worth more than raw raster speed in many cases. Look at the DLSS and FSR guide for details.

Finally, use balanced comparisons. A card that averages 5% higher FPS but has worse 1% lows might feel slower. Always check frame time consistency. If you plan to play at lower resolutions, a card with aggressive frame generation might give you smoother motion than a more powerful but older card. The best choice is the one that delivers the experience you want at the resolution you use, not the one with the prettiest bar chart. For a starting point, browse our best 16GB graphics cards.

What to pick for your use

Questions

What does 1% low FPS mean?

The 1% low is the average FPS of the slowest 1% of frames during a benchmark. It shows how well the card handles sudden frame drops. A high 1% low means smoother gameplay, while a low value indicates stutter.

Should I trust synthetic benchmarks like 3DMark?

Synthetic benchmarks are useful for controlled comparisons, especially for overclocking or testing. But they do not reflect real game performance, so use them as a supplement, not the deciding factor.

How do DLSS and XeSS affect benchmark numbers?

Upscaling technologies render at a lower resolution and use AI to reconstruct the image, boosting FPS. Frame generation creates synthetic frames. Benchmark numbers with these features are higher but not raw raster performance, so compare them separately.

Why do benchmark results vary between reviews?

Differences in test hardware, driver versions, temperatures, and even game updates can change results. Always compare numbers from the same source that tested multiple cards under identical conditions.

What is a good FPS target for gaming?

For most games, 60 FPS is playable, 144 FPS is great for fast action, and 240 FPS is for competitive esports. The right target depends on your monitor's refresh rate. Use benchmarks to see if a card can sustain your target.