On December 5, 1995, IBM took the wraps off its Deep Blue prototype, a supercomputer designed to beat the world’s greatest chess players. IBM would manage to achieve its goal two years later, after a host of software and hardware revisions. In 1997, Deep Blue famously triumphed over an at-his-peak chess Grandmaster Garry Kasparov, during a rematch in New York City. The win was a turning point for IBM, who was increasingly characterized as a has-been, with a dire share price to match. It was also a cornerstone in the company's approach to computing, pivoting from mere chunks of hardware to ‘thinking systems.’
Interestingly, Deep Blue originated from work on a chess chip, which started a decade earlier at Carnegie Mellon University. That hardware research project was dubbed Deep Thought, which will tickle The Hitchhiker’s Guide to the Galaxy fans.
1995: Losing versus a game running on a Pentium 90
Millions of moves per second sounds like a lot of brute force to crack a chess nut. However, it wasn’t yet enough to beat the best human players, nor even the best chess gaming computer programs of the era.
At its first outing at WCCC in 1995, Deep Blue would lose a decisive match against computer chess program Fritz running on a Pentium 90 PC. The loss showed IBM’s brute force technique couldn’t easily roll over Fritz’s well-curated opening book of game moves, positional heuristics, and chess knowledge.
Chess Grandmaster Kasparov faced Deep Blue for the first time in 1996. This time it showed it could do somewhat better – the IBM machine won Game 1 against the reigning world champ. After its promising start, the tables turned, though, with Kasparov triumphing 4-2.
1997: IBM brings moar brute force
IBM exacted its revenge in 1997, with Deep Blue victorious in a marathon 6-game rematch. Chess Programming says that the win was marginal, at 3½-2½ in favor of the machine. Some sources say that Kasparov accused IBM of cheating and demanded a rematch – an invitation that wasn’t accepted by Deep Blue, once it had grasped the headlines. IBM quotes the chess champ as grudgingly admitting, “I have to pay tribute, the computer is far stronger than anybody expected.”
The 1997 Deep Blue build was quite a significantly beefed-up build compared to the prototype specs we sketched out earlier. In this version of Deep Blue, the 1997 chess challenger was built using 30 workstation nodes of PowerPC processors controlling 16 chess chips each. IBM’s blog says this breakthrough design could “evaluate 200 million chess positions per second, achieving a processing speed of 11.38 billion floating-point operations per second. [FLOPS]”
Looking back at Deep Blue from the AI era
In 1997, we perhaps saw the first real signs of machines being able to rival the power of human thought and intuition – admittedly using a very different technique for success.
Fast-forward to the AI era, and as we approach the end of 2025, we have to comment on the fact that the AI-LLM industry is still pretty bad at chess. As the rapid pace of AI data center buildout and adoption continues, Deep Blue is a prescient reminder of where it all came from. We can only hope that there will be some RAM left over for the rest of us.
Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.
