Is Deep Blue source code open?

[tsoj]

Hi, I am just wondering if IBM released the source code for Deep Blue, I see no reason why they wouldn't.
I know it won't run on a desktop machine but it would be kinda nice to read through it.

[jp]

I am just wondering if IBM released the source code for Deep Blue, I see no reason why they wouldn't.

I can see a reason why they wouldn't. Someone could translate it and then we'd all see just how weak it is as software.

[Dann Corbit]

Deep Blue is not like most other chess programs.
It used Chess CPUs. Their instructions were things like "Rxc4" and the like.
The 480 chess CPUs were not just move generators, they also participated in the evaluation process.
They did not use much pruning because they were wary of overlooking tactics. Probably a good idea because Vincent's double null move check or Omid's null move verification had not yet been invented.
The Deep Blue machine analyzed 200 M nodes per second, with bursts as high as a billion.

But because it was a hardware solution (like Hitech and Hydra) the software would not do you any good. The software for Deep Blue was stuff to run on an IBM RS/6000 and the software was doing things like coordinating the work of the chess CPUs, which nobody has. So what would we do with such software?

Besides which, chess software grows exponentially stronger over time. While there are probably things we could learn from the software, it is really not generally applicable to the modern approaches. And current software is probably much stronger. So, for instance, the machine at the top of the list here:
http://www.ipmanchess.yolasite.com/amd- ... -bench.php

Code: Select all

Nodes/second   CPU                                                           Cores/Threads  BMI2/popcnt     Member
431.403.814     8x Intel Xeon Platinum 8168 8x24c/48t        384threads        BMI2                 noobpwnftw

Which is used by a member of this forum, would beat the ever loving pants off of Deep Blue. Deep Blue was a historical milestone and proof of that is that we still talk about it 20 years later and that only now is the exponential growth of the power of hardware surpassing that titanic monster.

[Jorge Garcia]

The evaluation was released and is from open domain

[MikeB]

Article was written by Feng-hsiung Hsu, the mastermind behind Deep Blue.

http://www.csis.pace.edu/~ctappert/dps/ ... s-deep.pdf

The movie "Game Over - Kasparov and the Machine" (2004) is a free stream on Amazon Prime. It actually got decent reviews , 60% by top critics, 76% by average joe critics.

[Werewolf]

Deep Blue was a historical milestone and proof of that is that we still talk about it 20 years later and that only now is the exponential growth of the power of hardware surpassing that titanic monster.

Well put. It was indeed a titanic monster and I hope one day some company will have the vision to take code like Stockfish, develop ASICs around that code and produce a new monster. But I doubt it will happen.

[Dann Corbit]

Deep Blue was a historical milestone and proof of that is that we still talk about it 20 years later and that only now is the exponential growth of the power of hardware surpassing that titanic monster.

Well put. It was indeed a titanic monster and I hope one day some company will have the vision to take code like Stockfish, develop ASICs around that code and produce a new monster. But I doubt it will happen.

I think we have proof that this is a mistake.

Developments like Hydra and Deep Blue are enormously expensive (millions of dollars).

The result of such technological developments is something that is a small constant faster than existing systems (e.g. 4x faster or 8x faster).
But since hardware and software both improve exponentially in a year or two, off the shelf stuff will equal it, and in twenty years it is no longer useful.

Consider the new AMD 7 nano-meter stuff. The chips will have up to 64 cores and 128 threads on a CPU unit, and there are lots of configurations where you can put two of these in a box. So 256 threads. It will probably set you back 12K just for the CPUs, but compare that with the cost of custom hardware like Deep Blue or Hydra. The finished system could be under 50K, fully loaded for chess. I am guessing well over 400M NPS.

Besides that, you have a system that is generally useful for anything else you want to do. A hardware chess machine can only play chess.

Custom hardware for chess is the wrong approach. Unless you want to be the first ever to beat the chess world champion and gain a ton of publicity, but we're too late for that.

Now, if you've got a billion dollars and you're bored and happen to like chess a lot, then have at it. Just know that the moment the machine is completed it is aging and in a few years it will be a very large and expensive paperweight.

[Robert Pope]

The evaluation was released and is from open domain

Where was that? I am not aware of the actual evaluation being released anywhere beyond the architecture layout discussing the evaluation terms they had available.

[jp]

Custom hardware for chess is the wrong approach. Unless you want to be the first ever to beat the chess world champion and gain a ton of publicity, but we're too late for that.

Now, if you've got a billion dollars and you're bored and happen to like chess a lot, then have at it. Just know that the moment the machine is completed it is aging and in a few years it will be a very large and expensive paperweight.

Yep. As software, Deep Blue was probably not even the best engine at the time. Junior maybe was.

[Werewolf]

Deep Blue was a historical milestone and proof of that is that we still talk about it 20 years later and that only now is the exponential growth of the power of hardware surpassing that titanic monster.

Well put. It was indeed a titanic monster and I hope one day some company will have the vision to take code like Stockfish, develop ASICs around that code and produce a new monster. But I doubt it will happen.

I think we have proof that this is a mistake.

Developments like Hydra and Deep Blue are enormously expensive (millions of dollars).

The result of such technological developments is something that is a small constant faster than existing systems (e.g. 4x faster or 8x faster).
But since hardware and software both improve exponentially in a year or two, off the shelf stuff will equal it, and in twenty years it is no longer useful.

Consider the new AMD 7 nano-meter stuff. The chips will have up to 64 cores and 128 threads on a CPU unit, and there are lots of configurations where you can put two of these in a box. So 256 threads. It will probably set you back 12K just for the CPUs, but compare that with the cost of custom hardware like Deep Blue or Hydra. The finished system could be under 50K, fully loaded for chess. I am guessing well over 400M NPS.

Besides that, you have a system that is generally useful for anything else you want to do. A hardware chess machine can only play chess.

Custom hardware for chess is the wrong approach. Unless you want to be the first ever to beat the chess world champion and gain a ton of publicity, but we're too late for that.

Now, if you've got a billion dollars and you're bored and happen to like chess a lot, then have at it. Just know that the moment the machine is completed it is aging and in a few years it will be a very large and expensive paperweight.

Yes I realise this, however, although it is far more practical (and cost effective) to make an AMD Rome system, the appeal of an ASIC based machine for a company may have some new grounds:

Where next after 7nm? How long will it take to double Rome's performance? Eventually we will not be able to shrink the die any more and at that point doubling of CPU performance will take many, many years.

All ASICs turn into paperweights, granted. But these days if we produced an ASIC machine 100x faster than your dual socket $50,000 AMD Rome machine above, it would take very long time for it to turn into a paperweight