Results
Leaderboards
Every board ranks inside a device class. A laptop never races a workstation card.
Boards
| Board | Ranks | By |
|---|---|---|
| Device class | Best verified run per operator in one class | MEM Score |
| Daily | Runs from the current UTC day | Position inside their own class |
| Weekly | Runs from the current season week | Position inside their own class |
| Efficiency | Classes with a single published memory configuration | Measured read ÷ published peak |
Class position
Top % = ⌈ (runs above + 1) × 100 ÷ runs in class ⌉, clamped to [1, 100]
- A class needs 5 verified runs before percentiles appear. Before that the result shows how many are missing.
- Daily and weekly boards only include classes with 5 distinct operators, so one person cannot open a class alone.
- Unranked runs are placed as a hypothetical extra entry and never exceed 100 %.
- Ties resolve by time, never by raw speed.
Why classes
Raw scores mostly measure the price of the card. Ranking inside a class measures what an owner controls: drivers, cooling, memory clocks, background load. An integrated GPU at the top of its class outranks a flagship in the middle of its own on the daily and weekly boards.
Boards are served from /api/leaderboard and cached for 10 seconds at the edge.