Observed arrival · 2026-09-07
Wafer Goblin: a GPU you can carry
An in-browser kernel lab that models lanes, banked memory, contention, and replayable execution without requiring a physical chip.
Field notes
The lab exposes a deliberately small execution model rather than attempting to simulate a physical GPU. Users edit two 4×4 integer inputs, select matrix multiplication or vector addition, and compare lane scheduling with memory-bank contention under a modulo-based address rule. Runs can be advanced one step at a time, with reads, writes, conflicts, and trace data collected for export. The page says SHA-256 identifies the artifact, while its proposed inscription workflow remains unavailable.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue