GPU Benchmarks calculations
Specific GPU performance is simply all the available benchmarks for this specific GPU. It might includes multiple systems like laptop models, equipped with various CPU models. At this point, all benchmarks are conducted using Windows OS. Graphics drivers are the latest available, unless there is no option to update. Linux gaming and synthetic benchmark may be added in the future.
Benchmarks include games and synthetic benchmarks. Gaming benchmarks are built-in or manual benchmarks of a software that its main purpose is a 3D game. Synthetic Benchmarks mean that the benchmark is a built-in, no associated directly with a game. It doesn’t mean that it’s not a part of a real-world in use software (like Luxmark).
A GPU benchmarks page (example) includes its raw specs, performance scores for available benchmark suites (read below), sometimes a description and gaming+synthetic benchmarks. It’s divided into application blocks (Bioshock Infinite, Battlefield 1, etc.) and each of them shows the worst, average and top average FPS performance (not the min/average/max FPS of a single laptop), across all systems tested for a specific resolution and graphics settings level. For example, if laptop#1 got 50FPS on average and laptop#2 got 70FPS, then worst=50, average=60 and top=70. At this point, there is no access to the min and max FPSs.
Benchmark Suite description
In order to easily compare GPUs gaming performance and to reflect more precisely – compared to a simple average – the 3D potential of a GPU, it is useful to have some kind of scoring system. So, let’s define a benchmark suite and say it’s a batch of benchmarks, meaning, a list of application at certain resolution and graphics settings. For example (1) Battlefield 1, Highest@1080p and (2) Crysis 3, Highest@1080p (but can it run Crysis?).
We can define multiple such suites for higher and lower 3D power. By choosing specific configurations, we can make it less likely to include stuff that can bias average performance to the wrong place. For example, World of Tanks, Dota 2 (at least in early versions) performance was bottlenecked with GPUs like GTX 1060 and up and CPUs like I7-6700HQ and I7-7700HQ. A simple average would show closer performance of a GTX 1070 to 1060 than it really is for well optimized demanding 3D games.
The problem cannot be negated completely, ofcourse, but that’s usually a better measure.
GPUs Comparison Calculations
A GPUs comparison page (GTX 1070 vs GTX 1060, for example) includes a description, quick performance comparison based on available benchmark suite, specifications table for both GPUs, simple average across all games and the detailed performance differences for each application. Comparison based on a benchmark suites probably reflects the differences more accurately in most cases, but the total average is there to give another indication. Comparison can be found via the “compare gpus” page.
Difference numbers for each application or game benchmark are calculated by taking the two highest (more demanding) shared graphics settings, done for each resolution. The result is shown for each resolution and graphics level (for example, 1080p and “high” graphics settings). Details are available for each application.
The total average differences are calculated as a simple average across all games or other application types, again separated by resolution. Median, mean and standard deviations are also calculated and shown to give a sense to what degree the results vary. These are only rough indicators, but are valuable.
Again, it is important to remember that the benchmarks often contain various types of systems in terms of hardware or software like an older or slower CPU, slower RAM, motherboard speed differences, MXM or soldered GPU, different drivers, OS and application version. So numbers shouldn’t be taken as an absolute, final performance indicators, but to give a rather accurate general performance range. If 10 benchmarks over various systems show a 9-11FPS on GPU “X”, you’d probably won’t get significantly better than that.
The separation to resolutions is important because different GPUs/systems shine mostly in specific situations. lack of VRAM, for example, can bottleneck the 3D performance in higher resolutions so the GPU is less utilized, giving a false notion for lower resolutions.
General Performance Score Calculations
Each GPU is assigned a performance score. This score is a very rough estimation of performance, but it allows comparing GPUs gaming performance easily across brands and architectures. Latest GPUs’ scores can be seen here.
The score is composed from average of two selected benchmark suites (see above). Not all GPUs have the required benchmarks, so in some cases the performance is estimated by simply comparing available benchmarks between two GPUs.