Kasparov vs. Deep Blue. Part I: Black Box

I continue a series of materials about the opposition of Garry Kasparov with computers, started by valemak . His articles, opening the topic, you can read here and here .

Despite the fact that by the mid-90s, chess programs for the PC had become difficult opponents for grandmasters, they still could not get closer to the level of world champion. Their chess strength in the classic time controls ranged from 2300 to 2400 rating points, which was still below the level of grandmasters (from 2500), and even more so, below the level of Kasparov's game (2800). In the future, many experts expected to improve the quality of the game of chess machines, but what could be done right now?

There were two ways here - to improve the quality of the chess programs themselves or to improve the hardware. I must say that over the previous years of development of chess programs, both methods were tried, and both gave a tangible gain - the strong code outperformed the weak, and switching to a fast machine, compared to a slow one, gave a significant increase in the strength of the game. But improving the quality of the code of chess programs is a relatively slow process, whereas more powerful computers actually existed. Thus, by attracting certain funds, it was possible to significantly increase the power of the game programs and, apparently, come close to the level of Kasparov. A high speed could give the desired increase of 300-400 rating points, even despite the existence of some pitfalls with this approach. And let the computer objectively be slightly inferior to Kasparov in strength, the chances of winning the match will still remain real. And the question is whether the computer of the best representative of humanity is stronger or not, while it is not so important - the essence of the match in reaching the peak, that is, opportunities at least once to jump to the very top, which at that time was the level of world champion among people.

Thus, only big cars at that time had real chances to cope with Kasparov. And such cars existed. That is why IBM has become interested in this issue. And became interested in the late 80s.
')
Winning the world chess champion would certainly be a sensation. Especially with good advertising. IBM could only pick a team of developers with the right experience on this issue. And such a team existed - it was the group that created the Dip Sot chess machine. This choice was also good because its developers focused on a fast machine, which exactly corresponded to the spirit of the company itself. However, the software component of the creators of Deep Sot was also quite on the level, otherwise no iron would save them.


Team Deep Blue, from left to right: Joe Hohen - soft; Grandmaster Joel Benjamin - debut, evaluation; Jerry Brody - tech support; Fen Xiong Xiu - hard; CJ Tan - the boss; Murray Campbell - software, evaluation.

I must say that IBM from the very beginning aimed at this team. It was the great success of the developers of Deep Sot, a machine that came close to the boundary of the grandmaster's strength of the game, and beat several grandmasters on long time controls — it led some people in the company to the idea of ​​creating an electronic champion.

Deep Sot is a chess computer created at the end of the 80s by a group of postgraduate students at Carnegie Mellon University, assembled literally on the knee - “considered chess” faster than many supercomputers of that time through the use of specialized ASIC iron. Of course, according to modern standards, its speed - and this is about 700 thousand positions per second - is not so great, and with some reservations it can be compared with the speed of a good modern smartphone game. But at that time, Deep Sot was far ahead of the same PCs in terms of the speed of chess calculations, and until the mid-90s, as a rule, easily beat them. And already in the late 80s, he was generally a vivid example of how much increase in the strength of the game could give an advantage in the speed of the hardware.

Thus, at the turn of 1989-90, having behind IBM all the financial strength, the Deep Sot team began developing the Deep Blue chess supercomputer. The power of Deep Blue was to consist of specialized microprocessors combined into a large system of parallel computing. Each of these microprocessors, even separately, could conduct chess calculations with a speed unprecedented for those times. When they were made, they were able to sort out about 2 million chess positions per second. Which roughly equals, albeit with certain reservations, the speed of the game on one core of modern mass processors from Intel and AMD.


One of the 1997 Deep Dip expansion boards, with chess processors (there are 8 of them)

Such a high speed, despite the old production process, was achieved due to the high specialization of integrated circuits. Such processors could count chess, and only chess. But then, unlike universal processors, they could do it very quickly. These were ASIC chips. They were managed by the “uncle” - an ordinary IBM RS / 6000 SP serial computer. It is his performance - 11.38 Gflops - that can often be found as an estimate of the speed of Deep Blue. However, this figure is only misleading, since chess calculations were carried out mainly on special processors, while the IBM server performed only a preliminary search and carried out the general management of the process. However, the speed of the server also directly influenced all calculations, because when distributing chess positions to special processors, the head computer found itself on the position of the bottleneck and became the bottleneck of the entire system.

Meanwhile, as time went on, and by 1995 the end of the development of Deep Blue seems to have already been seen. IBM decided that the time had come, and they sent a challenge to Kasparov for the match. Kasparov, of course, raised the abandoned glove. Especially as the proposed prize fund was quite consistent - after all, 500 thousand dollars for just 6 games.

Kasparov was not afraid of cars, and enjoyed playing with them. By that time, he had already raised his rating to record heights, won the match for the world championship with Anand, and played many games with computers. It remained only to fight the best of the cars. There was only one problem - he had to play with a supercomputer, which, like many other supercomputers, existed only in a single copy. The computers that Kasparov possessed could not even compare with Deep Blue. But the developers of Deep Blue, even with all their desire, could not help Kasparov before the match. The creation of a chess machine was reminiscent of all hands. Because of the need to refine the special processors, their final version was received from suppliers only at the very end of 1995. There was a general adjustment and adjustment of the machine. The creators of Deep Blue barely fit into the deadlines indicated in the contract for the match. But, in fairness, even if there was enough time, hardly anyone would allow Kasparov to conduct reconnaissance before the match in a future rival's “creative laboratory”. This is not practiced between people. Not intended here.

Despite the fact that quite a large number of people were engaged in developing a chess computer, and for quite a long time, the creation of Deep Blue, as is usually the case, was completed at the very last moment. As a result, there is no time left for high-quality opening preparation, and catching bugs. But to shift the timing of the match was already unrealistic.

The match between Kasparov and Deep Blue was scheduled to be held in Philadelphia. It was timed to celebrate the 50th anniversary of ACM . However, Deep Blue himself decided not to touch his place and left in New York. But just in case, IBM brought a couple of smaller computers to Philadelphia so that in case of problems with communication, you could quickly switch to them and at least somehow lead the party.

So, in the first days of February 1996, future rivals (and in the case of Deep Blue, its creators) came to Philadelphia to play a match that history hasn’t yet known ...

More information about the device Deep Blue can be found in Wikipedia or articles of the developers of the machine on the links:

• Chess Processors (English)
• General Description Deep Blue (English)

To be continued…