Mini PC Benchmarks

Static benchmark comparison page for mini PCs, designed to be hosted on GitHub Pages.

Files

• index.html - main page markup
• styles.css - page styles
• app.js - client-side logic for loading data, rendering charts/table, and saving column visibility
• devices.json - benchmark dataset consumed by the page
• device-links.json - editorial links keyed by device id

GitHub Pages

1. Push this folder to a GitHub repository.
2. In GitHub, open Settings > Pages.
3. Set the source to deploy from your default branch.
4. Open the published URL after Pages finishes building.

The page fetches devices.json and device-links.json at runtime, so it should be served over HTTP or HTTPS. Opening the HTML directly with file:// will usually fail because browsers block local fetch() requests.

Local Preview

Linux / macOS

Run a simple static server from this folder:

python3 -m http.server 8123

Then open:

http://127.0.0.1:8123/

Windows

No Python or Node required. Run the included PowerShell script (as Administrator):

.\serve.ps1

Then open:

http://localhost:80/

Note: Port 80 requires an elevated PowerShell session. Alternatively, change the port in serve.ps1 to anything above 1024 (e.g. 8123) to run without Administrator privileges.

Data Format

devices.json must contain an array of device objects. Example:

[
  {
    "id": "example-device",
    "name": "Example Device",
    "cb23s": 2000,
    "cb23m": 15000,
    "gb6s": 2800,
    "gb6m": 14000,
    "watts": 80,
    "handbrake": 90,
    "firestrike": 9000,
    "timespy": 3500,
    "noise": {
      "idle": 29,
      "load_default": 40,
      "load_performance": 45
    },
    "power_idle_watts": 8
  }
]

device-links.json stores links separately and is keyed by the same stable device id:

{
  "example-device": [
    {
      "label": "Affiliate link",
      "url": "https://example.com/buy",
      "kind": "affiliate"
    },
    {
      "label": "YouTube review",
      "url": "https://youtube.com/watch?v=example",
      "kind": "youtube"
    }
  ]
}

Add a New Mini PC Entry

Use this workflow each time you review a new device.

1. Open devices.json.
2. Add one new object inside the top-level array.
3. Add the same id key to device-links.json with either [] or a list of links.
4. Keep field names exactly as shown below.
5. Save the files and refresh the page.
6. Check table view, chart view, search, and the device details popup.

Copy/Paste Template

{
  "id": "brand-model-cpu",
  "name": "Brand Model CPU",
  "cb23s": null,
  "cb23m": null,
  "gb6s": null,
  "gb6m": null,
  "gbai_cpu": null,
  "gbai_gpu": null,
  "watts": null,
  "handbrake": null,
  "h264": null,
  "av1": null,
  "av1_hw": null,
  "firestrike": null,
  "timespy": null,
  "steelnomad": null,
  "coding": null,
  "photoshop": null,
  "premiere": null,
  "storage": null,
  "wireless_audio": null,
  "cpu_temp": null,
  "ssd_temp": null,
  "volume": null,
  "noise": {
    "idle": null,
    "load_default": null,
    "load_performance": null
  },
  "power_idle_watts": null
}

{
  "brand-model-cpu": []
}

Field Reference

• id: Stable slug used to map links and identify the device internally. Keep it unique and do not reuse it for another product.
• name: Device label shown in table/cards/charts.
• cb23s: Cinebench R23 single-core score. Higher is better.
• cb23m: Cinebench R23 multi-core score. Higher is better.
• gb6s: Geekbench 6 single-core score. Higher is better.
• gb6m: Geekbench 6 multi-core score. Higher is better.
• gbai_cpu: Geekbench AI CPU score. Higher is better.
• gbai_gpu: Geekbench AI GPU score. Higher is better.
• watts: Maximum power draw from the wall under load. Lower is better.
• handbrake: H264 encode time in seconds. Lower is better.
• av1: AV1 software encode time in seconds. Lower is better.
• av1_hw: AV1 hardware encode time in seconds. Lower is better.
• firestrike: 3DMark Fire Strike score. Higher is better.
• timespy: 3DMark Time Spy score. Higher is better.
• steelnomad: 3DMark Steel Nomad score. Higher is better.
• coding: Coding benchmark score. Higher is better.
• photoshop: Photoshop benchmark score. Higher is better.
• premiere: Premiere benchmark score. Higher is better.
• storage: 3DMark Storage Benchmark score. Higher is better.
• wireless_audio: Wireless Bluetooth audio benchmark score. Higher is better.
• cpu_temp: Maximum CPU temperature under load in C. Lower is better.
• ssd_temp: SSD temperature under load in C. Lower is better.
• volume: Chassis volume in liters. Lower is better.
• noise.idle: Fan noise at idle in dB(A) at 30 cm. Lower is better.
• noise.load_default: Fan noise under load in default profile. Lower is better.
• noise.load_performance: Fan noise under load in performance profile. Lower is better.
• power_idle_watts: Idle power draw from the wall. Lower is better.

Link File Reference

• label: Link text shown in the popup.
• url: Full outbound URL.
• kind: Short type tag. Current UI has built-in icons for affiliate and youtube.
• Link order is preserved exactly as written in device-links.json.
• If a mini PC has no links yet, use an empty array: "device-id": [].

Source Data Workflow

There are two PowerShell scripts for working with the raw CSV files in source:

• ./process-source.ps1 updates devices.json by resolving source labels to canonical device names.
• ./transpose-source.ps1 exports a flat CSV for inspection without any name matching or consolidation.

Import into devices.json

Use the importer script to map benchmark values from the source folder into devices.json:

./process-source.ps1

Optional parameters:

./process-source.ps1 -SourceDir ./source -DevicesPath ./devices.json -AutoAddDevices $false

• SourceDir: path to the source CSV folder (default: ./source)
• DevicesPath: path to devices.json (default: ./devices.json)
• AutoAddDevices: if $true, automatically create new device entries for unresolved source labels; if $false (default), add unresolved names to the report (default: $false)

The importer reads the benchmark CSVs, selects the preferred row per file, resolves each source device label, and writes the merged metrics back into devices.json. When AutoAddDevices is enabled, new devices are added with all metrics set to null and populated as metrics are imported. The importer also ensures every device has a stable unique id.

Resolver Order

process-source.ps1 resolves source labels in this order:

1. alias: exact match in the $aliases table.
2. canonical: normalized source name matches a normalized canonical device name already in devices.json.
3. fuzzy: token overlap score is high enough to auto-resolve the device.
4. unresolved: no safe match was found.

Name Normalization Rules

The resolver normalizes names before canonical and fuzzy matching:

• lowercases the label
• removes parenthetical suffixes such as (Gen4)
• removes storage sizes such as 512GB and 1TB
• removes generation tags such as Gen3, Gen4, and Gen5
• splits on non-alphanumeric characters
• removes noise words such as SSD vendor names and storage units
• keeps unique tokens only

Fuzzy Matching Rules

Fuzzy matching is intentionally conservative:

• at least 2 normalized tokens must overlap
• score is intersection / max(rawTokenCount, deviceTokenCount)
• score must be at least 0.34
• if both names contain identity tokens, they must overlap

Identity tokens are normalized tokens that contain digits and are at least 3 characters long. This helps prevent false positives such as matching unrelated devices that happen to share a brand or generic model family.

Canonical Name Preference

If multiple entries normalize to the same canonical key, the importer prefers the cleaner base name. It penalizes names with:

• parenthetical suffixes
• generation suffixes
• storage suffixes
• extra length

This prevents variants such as storage-specific labels from overriding the intended canonical device name.

Importer Output

process-source.ps1 logs the result of the import run:

• Updated metric entries: number of metrics written into devices.json
• Fuzzy matches: auto-resolved source labels with their token score
• Alias matches: source labels resolved through the explicit alias table
• Unresolved source names: labels that still need either a new alias or a new device entry

Example output:

Updated metric entries: 746

Fuzzy matches (auto-resolved using token scoring, score >= 0.34):
  'Minisforum M1 Pro (Gen4)' -> 'Minisforum M1 Pro-125H' (score: 0.75)

Alias matches (explicit mappings):
  'Minisforum M1 Pro 1TB Kingston (Gen4)' -> 'Minisforum M1 Pro-125H'

Export Raw Source Labels

Use the transpose script when you want to inspect the raw source labels without any matching logic:

./transpose-source.ps1

Optional parameters:

./transpose-source.ps1 -SourceDir ./source -OutputCsv ./source-transposed.csv

This writes a CSV with one row per raw device label and one column per metric. It preserves variant labels as separate rows, which makes it useful for auditing unresolved names, storage variants, and generation-tag variants.

New Device Naming Scheme

Use this canonical naming pattern for every new entry in devices.json:

Brand Model CPU

Examples:

• Beelink SER10 MAX HX 470
• ASUS NUC 15 Pro+ Ultra 9 285H
• GEEKOM IT15 Ultra 9 285H

Guidelines:

• Keep the full brand and model first.
• Keep CPU SKU at the end.
• Do not include storage size (512GB, 1TB) in canonical device names.
• Do not include SSD generation tags (Gen3, Gen4, Gen5) in canonical device names.
• Keep one canonical name per physical device model/CPU combination.

Resolve Unresolved Source Names

When process-source.ps1 reports unresolved names, use this decision flow:

1. If the unresolved name is a variant of an existing device, add/update an alias in process-source.ps1 ($aliases table).
2. If it auto-resolves correctly in the Fuzzy matches section, an alias is optional. Add one only if you want the mapping to be explicit and stable.
3. If it is a truly new device, add a new object to devices.json using the template above.
4. Add the same id to device-links.json, usually with [] first.
5. Run ./process-source.ps1 again and confirm the unresolved list is reduced or empty.

Alias Mapping Example

$aliases = @{
  'Source Label With Storage/Gen Suffix' = 'Canonical Device Name'
}

Handling Unresolved Names

Treat the unresolved list as a work queue for the current import run:

• add new devices for models you want to track in devices.json
• add aliases for labels that represent existing devices with extra storage or generation suffixes
• use ./transpose-source.ps1 to inspect the exact raw labels before deciding which path to take

Notes

• Use numbers (not quoted strings) for all benchmark, power, and noise values.
• For optional metrics with missing data, use null instead of 0.
• Keep devices.json valid JSON: commas between objects, no trailing commas.
• Keep device-links.json valid JSON and keyed by existing device ids.
• The app recalculates overall score and efficiency automatically from the numeric fields.
• Prefer aliases for ambiguous labels or labels you do not want depending on fuzzy-match behavior.
• Use ./transpose-source.ps1 to inspect the exact raw source labels before adding aliases.

Default Visible Columns

To change which optional columns are visible by default for first-time visitors, edit DEFAULT_VISIBLE_COLUMNS in app.js.

Visitors can change visible columns from the Columns button in the table toolbar. Their selection is stored in local storage under:

minipc-benchmarks.visible-columns

Use Reset defaults in the picker to restore the configured defaults.