GPUsed

How GPUsed GPU Comparison Works

Our GPU Comparison gives you a consistent place to start when deciding whether one graphics card is a worthwhile step up from another. The numbers describe GPUsed's continuous 1080p benchmark workload. They are not an estimate of the FPS you will see in a game.

Where the FPS figures come from

GPUsed supplied a model-level FPS result for each GPU shown in the comparison tool. GPUsed says its physical-card checks use a continuous FurMark run at 1080p, typically lasting 30 to 60 minutes while the card is checked for temperature and stability. The published figure is the supplied result for that GPU variant. We do not publish a result for a variant without a supplied FPS figure.

The historical count shown with a GPU is how many cards GPUsed says it has tested over time. It is not necessarily the number of individual runs used to calculate the displayed FPS. When a count covers a GPU family rather than one precise memory or BIOS variant, the tool says so. We keep missing counts separate rather than assigning them to a similar model.

How the comparison is calculated

The tool compares GPU X with GPU Y: (X FPS ÷ Y FPS − 1) × 100. A positive result means X recorded more FPS in this benchmark; a negative result means it recorded less. Reversing the cards changes the reference, so the two percentages are not exact opposites. Our rating uses a compressed performance scale: 100 × (GPU FPS ÷ highest GPU FPS) to the power of 0.35, rounded to one decimal place. The current highest result is 784 FPS and therefore scores 100. This makes lower and mid-range cards easier to distinguish on a /100 scale; it does not change the FPS or percentage difference, which use the original figures.

What this can and cannot tell you

A sustained graphics stress test is useful for comparing the supplied results under that workload. It does not measure a particular game, graphics preset, ray tracing, upscaling, frame generation, CPU limit or 1% low. Performance in your own games can differ, particularly between generations and driver versions. A model-level result also cannot establish whether a particular board fits your case or has the power connectors you need.

Use the benchmark difference as one part of an upgrade decision. Check the full model name, including Ti, Super, XT, XTX, VRAM and any memory or BIOS variant. Then compare the exact card's price, features, condition, dimensions and power requirements. Choose two GPUs in the comparison tool or browse published head-to-head comparisons.