Maximum Hash

[AdminX]

What are your maximum hash sizes?

Maximum Hash Findings

System Specs

Intel Core 2 Quad Q6700 @ 2.66 GHz
8GB Ram
Fritz 12 GUI
GUI Cache: 1 MB
Engine Cache: 10 MB

x86 Engines

2456 Hash Engines

Fritz 12

2431 Hash Engines

Deep Fritz 11
Deep Fritz 10
Deep Junior 7
Deep Junior 10.1
Crafty 23.00 (Chessbase Version)
Zap!Chess Zanzibar, Reykjavik, and Paderborn
Nimzo 8
Gambit Tiger 2.0
Chess Tiger 15.0
Hiarcs 8

2048 Hash Engines

Chess Tiger 2007 (Other Tiger Engines will use 2431 !?)
ProDeo 1.6
Hiarcs 12.1 (Odd that Hiarcs 8, 9, and Bareev will use 2431)
Movei 00_8_438
AnMon 5.66
Bright 0.4a
Ktulu 8
SmarThink 1.10
Spike 1.2
Yace 0.99.87 and Paderborn
List 512

512 Hash Engines

Ruffian 1.0.5: 512 (Will Crash with anything higher)


x64 Engines

4912 Hash Engines

Bison 9.6 (Wow, higher than the average)

4887 Hash Engines

Stockfish 1.5.1 JA
Naum 4
Deep Sjeng WC2008
Deep Shredder 12
Zappa Mexico II
Onno 1.1.1
Loop 13.6
Arasan 11.5

3656 Hash Engines

Rybka 3 (Will Crash with anything higher)
Note: Rybka's hash is odd, as I understand it uses 512 chunks, so max hash should be 3584 for 64-bit version)

[swami]

Interesting findings, Ted. There's room for creativity and originality in computer chess.

Does more hash contribute to better playing strength? I thought it doesn't and would only lead to crashes?

[Spock]

I set Zappa x64 with 6GB of hash once for a laugh, it and worked just fine

[Dr.Wael Deeb]

Thanks Ted for the interesting article
BTW,there is a newer version of Deep Toga than the one in your engine list:
Deep Toga 1.9.30
Dr.D

[Dr.Wael Deeb]

Interesting findings, Ted. There's room for creativity and originality in computer chess.

Does more hash contribute to better playing strength? I thought it doesn't and would only lead to crashes?

I do realy hope that you can make Omprakash believe in your words

http://rybkaforum.net/cgi-bin/rybkaforu ... ?tid=12708

[swami]

Interesting findings, Ted. There's room for creativity and originality in computer chess.

Does more hash contribute to better playing strength? I thought it doesn't and would only lead to crashes?

I do realy hope that you can make Omprakash believe in your words

http://rybkaforum.net/cgi-bin/rybkaforu ... ?tid=12708

I'm afraid I can't, Dr. I believe he has to experience it himself

[bob]

What are your maximum hash sizes?

Maximum Hash Findings

System Specs

Intel Core 2 Quad Q6700 @ 2.66 GHz
8GB Ram
Fritz 12 GUI
GUI Cache: 1 MB
Engine Cache: 10 MB

x86 Engines

2456 Hash Engines

Fritz 12

2431 Hash Engines

Deep Fritz 11
Deep Fritz 10
Deep Junior 7
Deep Junior 10.1
Crafty 23.00 (Chessbase Version)
Zap!Chess Zanzibar, Reykjavik, and Paderborn
Nimzo 8
Gambit Tiger 2.0
Chess Tiger 15.0
Hiarcs 8

2048 Hash Engines

Chess Tiger 2007 (Other Tiger Engines will use 2431 !?)
ProDeo 1.6
Hiarcs 12.1 (Odd that Hiarcs 8, 9, and Bareev will use 2431)
Movei 00_8_438
AnMon 5.66
Bright 0.4a
Ktulu 8
SmarThink 1.10
Spike 1.2
Yace 0.99.87 and Paderborn
List 512

512 Hash Engines

Ruffian 1.0.5: 512 (Will Crash with anything higher)


x64 Engines

4912 Hash Engines

Bison 9.6 (Wow, higher than the average)

4887 Hash Engines

Stockfish 1.5.1 JA
Naum 4
Deep Sjeng WC2008
Deep Shredder 12
Zappa Mexico II
Onno 1.1.1
Loop 13.6
Arasan 11.5

3656 Hash Engines

Rybka 3 (Will Crash with anything higher)
Note: Rybka's hash is odd, as I understand it uses 512 chunks, so max hash should be 3584 for 64-bit version)

Crafty will go as big as you want. I have run with 12G on our 16G cluster nodes. I have run much larger than that on other machines that occasionally have huge memory. I played in an old CCT event with a quad-dual-core AMD where each node (4 total) had 8 gigs of RAM. I tested with 24gb at the time (the largest it could do with 32 gigs of ram due to the power-of-2 way it allocates memory). I would not normally use anywhere near that however. My general rule of thumb is

hash = 16 * NPS * typical-time-per-move / 4;

16 bytes per entry, 75% of nodes are q-search where Crafty does no hashing. At one minute per move on the cluster node I use on ICC, this turns into

hash = 16 * 20M * X where X comes from the time control the opponent chooses. That is a reasonable upper limit. Lower will work just fine so long as you don't go too low.

[bob]

Interesting findings, Ted. There's room for creativity and originality in computer chess.

Does more hash contribute to better playing strength? I thought it doesn't and would only lead to crashes?

An interesting test is to do this:

Pick a reasonable position (middlegame). Set hash size to something very small, and run a 120 second search. Record the time taken. Double the hash size, and repeat. Continue until the time stops declining. Note that you may well find a size where the time jumps up a bit. Don't stop there.

That will give you an idea of how hash size affects the program during a game where a move takes about 2 minutes. You can repeat this for any time control you want to get a really good idea of the optimal value.