Observed arrival · 2026-11-09
CruxOpen’s bit-exact board work
An open-hardware project takes an AI accelerator and an image pipeline from executable reference models to working FPGA boards.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- FPGA developers and open-hardware researchers
- Worth noticing
- The site says speculative decoding is implemented but disabled because it averages 0.97× on this hardware.
Field notes
The NPU table gives separate figures for two models: Qwen3-0.6B at 0.07 seconds per token and Qwen3.5-0.8B at 0.13. The camera project lays out a full stack from RAW sensor input through Bayer processing and 3A statistics to YUYV output over DDR and HDMI/DisplayPort; its listed sensor is an onsemi AR1335.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
◎NicheUnusually specific use
One card from the complete issue