Is Deep Blue source code open?

[Jorge Garcia]

I remember glancing the code ten years back when I was helping Thomas gawk in improving toga eval, I have made a search in Google but I don't find the code, anyway is not worth it. You can find more interesting ideas in whatever up to date engine than in deepblue

[ker2x]

Unless you have infinite money like IBM or Google, it doesn't make sense to do an ASIC.
However, we're seeing more and more integration between FPGA and CPU (you can even rent some on Amazon Web Service), but i didn't think there is an FPGA powered chess engine that beat an X86_64 CPU... yet ?

[tsoj]

I thought it just has some historical value a little bit like the source code of the apollo guidance computer (https://github.com/chrislgarry/Apollo-11/).

[Milos]

Unless you have infinite money like IBM or Google, it doesn't make sense to do an ASIC.
However, we're seeing more and more integration between FPGA and CPU (you can even rent some on Amazon Web Service), but i didn't think there is an FPGA powered chess engine that beat an X86_64 CPU... yet ?

That's because ppl who can actually effectively program FPGA (not just use automated C to verilog/VHDL translators that are a joke) and know something about chess programming are almost non-existent or if they exist they earn far more with their daily job to have time to waste on such hopeless projects.
Btw. all the DeepBlue FPGA power combined is less than today's Virtex Ultrascale+ single FPGA like VU13P.

[mschribr]

While we are comparing IBM's deep blue hardware to today's hardware and deep blue chess performance to today's computer chess performance. It's also interesting to compare IBM 1997 to IBM 2017. In 1997, IBM earned 78.5 billion. In 2017, IBM earned 79.1 billion. In 20 years an increase of 0.8%. Not much. IBM is just treading water.

[Milos]

While we are comparing IBM's deep blue hardware to today's hardware and deep blue chess performance to today's computer chess performance. It's also interesting to compare IBM 1997 to IBM 2017. In 1997, IBM earned 78.5 billion. In 2017, IBM earned 79.1 billion. In 20 years an increase of 0.8%. Not much. IBM is just treading water.

That's pretty ridiculous statement. Take GM's or Ford's revenue in 1997 and 2017 and you are also gonna notice they are trading water, right?
Basically no company outside pure software or retail had more than 100% revenue gain in the last 20 years.

[yurikvelo]

AFAIK Deep Blue source code had evaluation function in a generalized form. This function was "trained" and this training data is not part of source code, but rather dynamic data, which can be enhanced with more games thrown on it and also manually intervented by team stuff.

I doubt DB can be reproduced even with sources and good emulator. It can be trained with new games dataset, but it's playing style and strength will be different.

[Dann Corbit]

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.

Well, 5 nm is making progress with the same company doing the 7 for AMD.
Of course, there is some limit because 1 atom wide traces are not practical.
I guess that a new technology will come along.
We went from relays to vacuum tubes to transistors to integrated circuits to parallel integrated circuits. Something will be "next".

ASIC and FPGA are both silicon too, so they suffer from the same limits.
What you gain is something a small constant better than compiled code.
I think a revolution in speed needs a new backdrop besides silicon.

[Milos]

AFAIK Deep Blue source code had evaluation function in a generalized form. This function was "trained" and this training data is not part of source code, but rather dynamic data, which can be enhanced with more games thrown on it and also manually intervented by team stuff.

I doubt DB can be reproduced even with sources and good emulator. It can be trained with new games dataset, but it's playing style and strength will be different.

Seriously man, where does all this BS come from. Trained function, really?
Also so much BS in this thread, really shocking. Just google it ppl for god sake, "Deep Blue", Murray Campbell, A. Joseph Hoane Jr., Feng-hsiung Hsu, Artificial Intelligence 134, Elsevier 2002. All is explained well and in detail (not like A0 crappy preprint).

[Ajedrecista]

Hello:

I remember glancing the code ten years back when I was helping Thomas gawk in improving toga eval, I have made a search in Google but I don't find the code, anyway is not worth it. You can find more interesting ideas in whatever up to date engine than in deepblue

I thought it just has some historical value a little bit like the source code of the apollo guidance computer (https://github.com/chrislgarry/Apollo-11/).

I fully agree with Jost (thank you very much for the reminder of Apollo!). It would be great to see how the eval looks like, if available, not for its usefulness but for its historical value.

IBM uploaded the logs of the Deep Blue - Kasparov match in its web and it is still reachable with the help of the WayBack Machine:

http://web.archive.org/web/200511251620 ... atch/html/

https://web.archive.org/web/20080701232 ... ml/c.shtml

I attach these logs in this post.

Regards from Spain.

Ajedrecista.