How We Test PC Hardware

The reference test bench

We use a consistent, current-platform bench for GPU and general gaming comparisons. The exact part model is named here so a result has a home, not just a chart.

For a GPU review, the card under test replaces the reference GPU. For CPU testing, the reference GPU stays in place unless the workload needs a different class of card. We list material departures—driver, BIOS, memory profile, or game patch—beside the affected result.

Benchmark methodology

Our game suite includes a mix of demanding current releases and repeatable staples: Cyberpunk 2077, Baldur's Gate 3, Forza Horizon 5, Hogwarts Legacy, Starfield, and Counter-Strike 2. We also use Blender Classroom, Cinebench 2024, and a sustained file-transfer workload when the part calls for it.

We test at the resolution and quality settings buyers are likely to use: 1080p, 1440p, and 4K where the hardware makes sense; high or ultra presets; and clearly labeled ray tracing, upscaling, and frame-generation settings. Each game gets a repeatable built-in benchmark or a fixed manual route after a warm-up pass. We run each test at least three times, discard an obvious invalid run, and report the arithmetic average FPS. We check 1% lows and frame-time behavior before calling a result smooth.

Cost per frame

The working formula is: cost per frame = price ÷ average FPS.

Price is the current street price or, when a product is newly launched, the stated launch price. Frames means the average FPS from the same game suite and settings. Lower cost per frame is better only after the hardware clears the resolution, features, noise, and upgrade target for the build. A bargain that needs a new PSU is not a bargain; we say so.

Thermals and stability

We test thermals in a normal airflow case with a documented fan profile, logging ambient room temperature and reporting sustained—not just first-minute—temperatures. CPU and GPU checks include a 30-minute combined gaming or productivity load, plus the workload-specific tests above. For stability, we look for crashes, visual artifacts, driver resets, clock behavior, and repeatable performance across the full set of runs. An overclock or undervolt that survives one screenshot does not earn a recommendation.

Fresh data and prices

Drivers, game patches, and prices move fast. We retest when a major driver, BIOS, game patch, or platform change is likely to alter a buying decision, and we date material updates. Price checks use current listings from major US retailers and manufacturer stores; we compare like-for-like models, exclude one-off marketplace listings, and flag rebates, bundles, and temporary sales. If a price is stale or unavailable, we label it rather than pretending the chart is current.

Want us to check a result? Send the full hardware list, settings, driver version, and your average FPS through our contact page.