Re: [computer-go] question regarding Hydra Chess Supercomputer

["chrilly" <chrilly@xxxxxxxx>]

>
>Do you believe that Deep Blue was able to utilize global transposition
>tables  throughout the  whole search?
>
As said before, the Hydra and DB architecture is very close. In case of
Hydra the software side is currently a Linux-Cluster of 8-Dual PCs (a
16-Dual-Cluster is under construction). The Duals are connected by a
Myrinet-Networks. On the software level communication is done by MPI.
Each CPU has a hardware "coprocessor" on a FPGA-card. Communication between
CPU and the card is done via the PCI-Bus (a severe bottleneck).
Hashtables are more or less useless. In a parallel system one must
distribute the hashtable-entries. If a card searches a given subtree, it is
very unlikely that this subtree is searched the next time by the same card.

A hashtable would only be usefull, if there would be a very fast inter-card
network such that the cards can exchange the hashtable information like the
software does over Myrinet. There are now some FPGA-clusters of this type
available. But then one gets the problem how to interface the sofware and
the hardware-cluster. Just plugging in a card in each PCI-slot is much
simpler and cheaper.
Sending the hash-information over the PCI-bus to the software side and then
doing the distribution is by some factors too slow. In fact even the fastest
networks in the current FPGA-Clusters are too slow for this purpose. One
would have to restrict oneselves to entries at the root of the hardware
search.

>  It's a shame he never got to develop  the chip further after the famous
matches...
>
IBM was not interested in chess (thats one big difference between IBM and
the Hydra-Sponsor). For them the marketing-effect was important. From the
marketing perspective IBM could only loose in another game. A loss would
damage the mythos. Feng Hsu did also not fit in the cooperate-culture of
IBM. He is a very direct guy.
At the beginning of the Hydra/Brutus project I was in the U.S. to get some
intro by Ken Thompson. I met also Feng Hsu.
At that moment I could not stand anymore the "positive thinking" of the
americans. Everything is "nice", "great". In Austria one says even to very
nice things "it is not too bad".  If someone tells you in the U.S. that he
left his company and got another job he says: "I looked for a new
challenge". Feng Hsu said: "The first day I entered IBM I knew that I will
not stay forever here".  I loved this sentence. Thats the Austrian way of
expressing things.

Chrilly


_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/