On February 10th, 1996, IBM's Deep Blue defeated reigning champion Garry Kasparov in a historic chess game. Deep Blue would later win a 3.5-2.5 rematch a year later, but the machine required several million dollars in research and 30 P2SC processors running at 120 MHz plus 480 custom VLSI chess co-processors. The upgraded Deep Blue could analyze over 200 million positions per second and was once the 259th most powerful supercomputer globally.
However, a modern 200 MHz Intel Pentium Pro from 1995 running a custom Stockfish engine estimated at 3300 Elo easily surpasses Kasparov's 2800. This proves commodity hardware with superior software can outperform specialized supercomputers. The article notes that +720 of the +1335 Elo gains in chess AI come from better programming, not faster hardware.
Two experiments demonstrate this: a 1996 Crafty engine on modern hardware reached 2614 Elo, and adding modern engine knowledge like NNUE to older code further increases strength. The best 1996 engines rated only 2300 Elo compared to Deep Blue's 2700, while modern Stockfish sits near 3700. The conclusion suggests most engine progress stems from improved move selection algorithms rather than raw computing power.
Source: Hacker News · Summarized by HeadlinesBriefing