January 29, 2026
8-bit or nah? Fight me
Playing Board Games with Deep Convolutional Neural Network on 8bit Motorola 6809
1978 ‘8‑bit’ chip plays Go with AI; comments erupt over 16‑bit vibes
TLDR: Researchers ran a Go‑playing neural network on a 1978 Motorola 6809, reaching strength similar to [GNU Go]. Comments split over whether the chip is truly 8‑bit or a 16‑bit hybrid, with jokes about “board games” meaning only Go—showing retro hardware can still host modern AI tricks.
A retro miracle just dropped: researchers made a 1978 Motorola 6809 run a Go‑playing AI (a convolutional neural network, aka pattern‑spotting brain) on a Thomson MO5. It even plays about as well as GNU Go. The community? Instantly split between nostalgia and nitpicking.
Strongest opinions: one camp insists the 6809 isn’t “really” 8‑bit. User pklausler swaggered in with “I think of the 6809 as a 16‑bit,” and the rebuttal brigade fired back with “hybrid processor, 16 on the inside, 8 on the bus” and a Wikipedia citation. Translation: semantics war over whether this retro flex truly counts as an 8‑bit win.
Drama level: spicy. Some readers rolled their eyes at the headline—“Board games” here means Go, full stop—sparking jokes about waiting for AI Settlers of Catan on a cassette tape. Others loved the “AI on a diet” vibe, guessing the model used tricks like quantization (shrinking numbers to fit tiny memory).
The mood swings between wow, grandpa chip can game and well actually, it’s 16‑bit-ish. But the crowd mostly agrees: squeezing modern AI into a museum‑grade machine is both ridiculous and glorious. Call it retropunk engineering—proof that you don’t need a monster GPU to get clever moves.
Key Points
- •The paper demonstrates a Go-playing convolutional neural network running on a Motorola 6809 8-bit CPU.
- •Implementation was performed on a Thomson MO5 microcomputer.
- •The system achieved playing strength comparable to GNU Go.
- •The work highlights that neural network inference can be efficient on small, low-power devices.
- •Published in Game Programming Workshop 2023 proceedings by the Information Processing Society of Japan (pages 66–69, 2023-11-10).