smatovic wrote: ↑Mon Oct 13, 2025 7:20 pmEd and Uri estimated +200 Elo for backporting new techniques onto 6502, and Bill Foster did his Sargon 1 port onto x86, 3 ply extra.
smatovic wrote: ↑Sat Oct 11, 2025 4:02 pm
[...]
I know, the Sargon port by Bill Foster:
That's not what Bill Forster did: according to the post you linked, he took the Sargon program, unchanged, and ran it on hardware which was 6000x faster.
Not sure we are talking about the same thing....
Converting the classic Z80 code to X86 with no emulation or simulation to allows us to really how quickly it can run on modern machines.
smatovic wrote: ↑Mon Oct 13, 2025 7:20 pmEd and Uri estimated +200 Elo for backporting new techniques onto 6502, and Bill Foster did his Sargon 1 port onto x86, 3 ply extra.
smatovic wrote: ↑Sat Oct 11, 2025 4:02 pm
[...]
I know, the Sargon port by Bill Foster:
That's not what Bill Forster did: according to the post you linked, he took the Sargon program, unchanged, and ran it on hardware which was 6000x faster.
Not sure we are talking about the same thing....
Converting the classic Z80 code to X86 with no emulation or simulation to allows us to really how quickly it can run on modern machines.
syzygy wrote: ↑Mon Oct 13, 2025 8:14 pm
It looks like Sargon did not implement the algorithmic improvement known as alpha-beta.
I did not look into the sources, but:
The first Sargon version had a 10x12 board board representation and a two ply alpha-beta search without quiescence but a SOMA like exchange evaluation [3] [4].
syzygy wrote: ↑Mon Oct 13, 2025 8:14 pm
It looks like Sargon did not implement the algorithmic improvement known as alpha-beta.
I did not look into the sources, but:
The first Sargon version had a 10x12 board board representation and a two ply alpha-beta search without quiescence but a SOMA like exchange evaluation [3] [4].
This version of Sargon used fixed depth, full width search. In other words it did not prune its search tree, considering every single move out to a fixed depth. The depth was a user selectable option, with 6 (ply) being the maximum setting. On my (cheap, aging) commodity laptop, sargon-engine.exe might take something like 30 seconds to perform the 6 ply search in a typical middlegame position. I expect this to be between two and three orders of magnitude faster than the original Z80 machine, so I don't imagine many users set the search depth to maximum!
6 ply in 30 seconds would mean about 24Mnps (with a branching factor of 30).
That could be right, although perhaps a bit high.
With alpha-beta the effective branching factor should be around perhaps 10, which would amount to 33Knps, which is way too low.
edit: src/sargon-minimax.cpp includes comments talking about alpha-beta, so now I am not sure.
syzygy wrote: ↑Mon Oct 13, 2025 8:14 pm
It looks like Sargon did not implement the algorithmic improvement known as alpha-beta.
I did not look into the sources, but:
The first Sargon version had a 10x12 board board representation and a two ply alpha-beta search without quiescence but a SOMA like exchange evaluation [3] [4].
This version of Sargon used fixed depth, full width search. In other words it did not prune its search tree, considering every single move out to a fixed depth. The depth was a user selectable option, with 6 (ply) being the maximum setting. On my (cheap, aging) commodity laptop, sargon-engine.exe might take something like 30 seconds to perform the 6 ply search in a typical middlegame position. I expect this to be between two and three orders of magnitude faster than the original Z80 machine, so I don't imagine many users set the search depth to maximum!
6 ply in 30 seconds would mean about 24Mnps (with a branching factor of 30).
That could be right, although perhaps a bit high.
With alpha-beta the effective branching factor should be around perhaps 10, which would amount to 33Knps, which is way too low.
edit: src/sargon-minimax.cpp includes comments talking about alpha-beta, so now I am not sure.
I do not know either. In the other post Foster mentions an BF of 20 which translates to 2.13Mnps for 6 ply in 30 seconds.
***edit***
"full width search" can be interpreted as no selective search present.
smatovic wrote: ↑Mon Oct 13, 2025 8:45 pm
I do not know either. In the other post Foster mentions an BF of 20 which translates to 2.13Mnps for 6 ply in 30 seconds.
***edit***
"full width search" can be interpreted as no selective search present.
Then probably it is alpha-beta but with rather bad move ordering (not having a TT indeed does not help).
smatovic wrote: ↑Mon Oct 13, 2025 7:20 pmEd and Uri estimated +200 Elo for backporting new techniques onto 6502, and Bill Foster did his Sargon 1 port onto x86, 3 ply extra.
smatovic wrote: ↑Sat Oct 11, 2025 4:02 pm
[...]
I know, the Sargon port by Bill Foster:
That's not what Bill Forster did: according to the post you linked, he took the Sargon program, unchanged, and ran it on hardware which was 6000x faster.
Not sure we are talking about the same thing....
Converting the classic Z80 code to X86 with no emulation or simulation to allows us to really how quickly it can run on modern machines.
For the fun of it one should remove A/B from your engine and watch the search depth, suddenly slow as a snail and the hundreds of elo loss that will come with it. I once tried it, the first 3 iterations without A/B in order to get a well ordered move list that could profit from iteration 4 an on.
No cigar, takes too long.
90% of coding is debugging, the other 10% is writing bugs.
smatovic wrote: ↑Mon Oct 13, 2025 7:20 pmEd and Uri estimated +200 Elo for backporting new techniques onto 6502, and Bill Foster did his Sargon 1 port onto x86, 3 ply extra.
smatovic wrote: ↑Sat Oct 11, 2025 4:02 pm
[...]
I know, the Sargon port by Bill Foster:
That's not what Bill Forster did: according to the post you linked, he took the Sargon program, unchanged, and ran it on hardware which was 6000x faster.
Not sure we are talking about the same thing....
Converting the classic Z80 code to X86 with no emulation or simulation to allows us to really how quickly it can run on modern machines.
Wow! Throughout that code, it is astonishing how much is being done with so few CPU instructions: mclane - take a look - the secret of the strength of the dedicated chess computers from the golden age is right here!
Human chess is partly about tactics and strategy, but mostly about memory
