We use game testing data to identify the components that determine game performance at each resolution and settings level.
GPU-limited games
The performance of most new games, especially at higher resolutions and visual quality settings, is bottlenecked by the GPU. The performance of the game is determined by the graphics performance level of the system.
3DMark benchmark tests like Port Royal and Wild Life measure only GPU performance. Other benchmarks, such as Time Spy and Fire Strike, include Graphics Tests that measure GPU performance and CPU/Physics tests that measure CPU performance.
The chart below shows the relationship between 3DMark Time Spy Graphics score and average frame rate in Fortnite. Each orange dot shows the performance of a system tested by UL.
The chart shows that a PC that scores 4,000 in the 3DMark Time Spy Graphics test can run Fortnite at 50 FPS on average at 1440p with the highest visual quality settings. A 3DMark score of 6,000 translates to an average of 70 FPS in Fortnite. A score of 12,000 translates to 140 FPS.
The charts below show the correlation between 3DMark Time Spy Graphics score and average frame rate in Apex Legends and Red Dead Redemption 2.
It’s these correlations that enable us to estimate game frame rates from 3DMark scores.
We check the correlation between game performance and benchmark scores for each combination of resolution and visual quality settings separately. Any anomalous results are retested before they are used in statistical modeling.
CPU-limited games
With older games, or when using a lower resolution and medium quality settings, the game frame rate may not scale with graphics performance. This is almost always a result of the CPU limiting the performance of the game.
To detect games that are bound by the CPU, our test matrix features systems that have different CPUs but the same GPU.
For CPU-limited games, we look for correlations between the game frame rate and CPU performance as measured by 3DMark CPU tests. As different games and game engines make use of CPU cores in different ways, this analysis is often more complex. It is performed separately for different CPU vendors and CPU core counts.
Games with complex bottlenecks
When estimating the performance of a game that can be limited by either the GPU or the CPU depending on the system, the performance levels of the CPU and GPU are estimated separately to determine the bottleneck. The lower of these two estimations is used to estimate the game frame rate for that specific configuration of CPU and GPU.
Estimating game performance for multi-GPU systems
For games that support AMD CrossFire or NVIDIA SLI, game performance scaling with multi-GPU systems is tested, checked and modeled separately. Game performance estimates for multi-GPU systems will reflect the use of multi-GPU when supported by the game. Most games do not benefit from having a second GPU in the system, however.
If a game does not support multi-GPU, the performance estimate is based on the equivalent single-GPU system.