GPU Testing Protocol

Each card receives an internal asset tag and a standardized visual inspection. Technicians document PCB condition, fans, I/O ports, power connectors, and evidence of liquid damage. Any severe physical damage is immediately failed and routed for repair assessment.

• Check PCB, solder joints, and power connectors for corrosion or damage.
• Inspect I/O bracket and display ports for deformation or burn marks.
• Photograph the serial label and add standardized asset images to the unit record.

A scripted identification run collects vendor and device IDs (WMI / lspci), queries GPU-Z SDK, and runs vendor tools (nvidia-smi, AMD ADL/AGS) to extract core identifiers. Results are cross-referenced against the Master Specification Database. Any mismatch triggers manual verification.

The goal: populate the GPU profile (core, AIB partner, memory type, clocks, TDP) and retrieve a serial number. When serials are not accessible via software (consumer GeForce limitation), a high-resolution macro photo of the sticker is logged and used for manual entry.

Using the composite hardware key (Vendor ID + Device ID + SUBSYS) the system attempts an automatic lookup into our canonical GPU specs database (sourced from TechPowerUp + internal audits). If multiple matches or ambiguities are returned, the unit is flagged for human review and manual schema reconciliation.

To ensure performance differences are attributed only to the GPU, we test on a standardized, documented bench. The bench components are fixed: Ryzen 7 7800X3D (or equivalent), 32GB DDR5, NVMe OS drive, 1200W ATX 3.0 PSU, open air chassis, and the latest certified drivers. Every test run logs the driver version, OS build, and a full HWiNFO sensor trace.

We run a curated suite of 12 games and workloads (various engines and APIs) across 1080p, 1440p, and 4K. The suite is updated annually to reflect modern engines and path-tracing workloads. For each title we collect average FPS, 1% low, 0.1% low, frame-time percentiles, and a captured video of the run for qualitative verification.

We run 3DMark (Time Spy, Port Royal where relevant), Unigine and industry-standard renderer tests for compute and ray-tracing workloads. For professional cards, specific workloads such as Blender and SPECviewperf are included.

Each synthetic run is performed multiple times; we report averages and the highest stable run while ensuring thermal stability. Run-to-run variance is calculated and displayed on the product data page.

Sustained heavy-load runs (30+ minutes) are used to evaluate thermal throttling and long-term performance curves. We log GPU junction (Tj), VRM thermals, fan RPM curves, and any clock-step behavior over time. These long-duration profiles reveal differences between peak and sustained performance — critical for buyers who run long sessions.

Board power is measured using an independent power meter (ElmorLabs PMD series) to bypass unreliable vendor APIs. We report typical and peak board power, and compute FPS-per-watt for key titles/resolutions to present an efficiency metric that matters to thermals, PSU selection, and long-term running costs.

All data (sensor logs, benchmark output, photos, and manual inspection notes) are assembled into a time-stamped, verifiable package. A senior technician reviews the full record and signs off before a unit is moved to certified inventory. Any deviations from the expected spec or performance baselines are documented in the Configuration Delta report attached to the product listing.

The product detail page (PDP) surfaces the key measured metrics: gaming scores (with 1%/0.1% lows), synthetic scores, power and thermal graphs, and the exact parts list + serial photo. We include explanatory tooltips to help buyers interpret the data and make an informed choice.