Game-AI Forum

While training deep-learning neural networks often requires considerable amounts of computing power, inference is efficient, and can be run on small devices. Cell phones are a typical example, but they are still rather powerful. The research presented in this paper takes the challenge to the extreme by running a Go-playing convolutional neural network on the 6809 CPU, an 8-bit microprocessor launched by Motorola in 1978. The software was implemented on a Thomson MO5 microcomputer, and reached a playing strength on par with GNU Go.