Starfield

Starfield is widely used as a hardware benchmark because, as reviewers repeatedly noted at launch, it is an unusually demanding rasterization game relative to how it looks. TechSpot benched 32 GPUs across three presets and three resolutions and highlighted how, for example, a Radeon RX 6800 XT managed only 38 fps at 4K Medium, 33 fps on High and 29 fps on Ultra — far below the 40-50 fps that same card averages in other modern titles — making the game a harsh, revealing stress test of GPU horsepower rather than of any single new rendering feature. Tom's Hardware similarly framed it as a title where hardware that normally sits well ahead only barely pulls away, calling the RX 6800 XT roughly the minimum for a smooth 1080p Ultra experience without upscaling.

Multiple outlets emphasize that Starfield is both GPU- and CPU-bound, which is part of why it makes a good all-around system benchmark. GamersNexus explicitly notes the game is taxing on both GPUs and CPUs, and produced a dedicated CPU bottleneck comparison in addition to its GPU testing. TechPowerUp observed that Starfield is very CPU intensive, especially inside large cities, where even a high-end GPU like an RTX 4080 can become CPU bottlenecked at 1440p and below — behavior that makes dense urban areas useful for isolating CPU and memory performance. Because the game launched as an AMD partner title, reviewers also used it to examine vendor-specific behavior: AMD Radeon cards performed unusually competitively against NVIDIA, which GamersNexus and Tom's Hardware both flagged as departing from typical rasterization scaling.

The game is a pure rasterization workload with no ray tracing, so it primarily stresses raw shading throughput, VRAM and CPU simulation/streaming rather than RT or path-tracing hardware. ComputerBase's technical test specifically broke out results by FHD, WQHD and UHD along with frametimes and VRAM usage, underscoring memory capacity as a factor. Upscaling is also central to the benchmarking story: at launch Starfield only supported AMD FSR 2.2, and TechPowerUp found its FSR implementation delivered smaller-than-usual gains (around 25% in Quality mode versus the ~40% typical of other FSR titles), so reviewers generally test at native resolution to keep results GPU-bound while still noting that upscaling is effectively required to hit high framerates.

Choosing a consistent, worst-case-but-common location is a key reason reviewers trust Starfield results. GamersNexus described spending hours exploring gameplay and environments — the tutorial cave, moon surface, ship, space flight, on-foot and space combat, and New Atlantis City — before settling on a repeatable test location, and confirmed the game has no built-in benchmark, so manual runs (with FSR, dynamic resolution and variable rate shading disabled to stay GPU-bound) are required for clean comparisons.