GPUbenchmarks

GPU benchmarks are standardized tests used to evaluate the performance of graphics processing units (GPUs). They provide repeatable measurements of rendering speed, raw compute capability, and stability under pressure, helping consumers, reviewers, and manufacturers compare architectures and configurations.

Benchmarks are broadly divided into synthetic tests and real-world benchmarks. Synthetic tests simulate representative workloads and render scenes or perform computations in a controlled environment. Real-world benchmarks measure performance using actual games or applications, providing results that more closely reflect typical usage. Common metrics include frames per second (FPS), frame-time distributions such as 1% and 0.1% low, and compute throughput in FLOPS or render pipeline metrics.

Popular suites and tests include 3DMark (Time Spy for DirectX 12, Fire Strike for DirectX 11, and Port Royal for ray tracing), Unigine Heaven and Valley, Basemark GPU, GFXBench, SPECviewperf, and PassMark. Some tests emphasize stability and power consumption, others stress tessellation, ray tracing, or AI-based tasks. In addition, many reviewers run real-game benchmarks at common resolutions (e.g., 1080p, 1440p, 4K) with preset quality levels.

Methodology matters. Test rigs should specify CPU, memory, drivers, operating system, cooling, and power limits, as these can affect results. Resolution, graphics quality, API (DirectX, Vulkan, Metal), and driver optimizations influence outcomes. Reproducibility is improved by running multiple iterations and reporting averages and lows.

Limitations include the fact that synthetic benchmarks may not predict real-game performance across engines and drivers. GPU benchmarks are most informative when used to compare similar GPUs under the same test conditions.