Grapefruit 1.0 is weaker than 1.0 beta

[Jouni]

You don't need hundreds of games to see that. It scored worse against R3, N4 and F10 in my test matches. Also in testsuites it's slower/weaker. So what's the story behind removal of "beta" tag? Note, that I tried a lot of different compiles without seeing any real speed difference.

Jouni

[Denis P. Mendoza]

You don't need hundreds of games to see that. It scored worse against R3, N4 and F10 in my test matches. Also in testsuites it's slower/weaker. So what's the story behind removal of "beta" tag? Note, that I tried a lot of different compiles without seeing any real speed difference.

Jouni

Good tests never lie! I also noticed. At least we know now. I haven't heard news from Vadim yet. But the readme stated many changes:

List of major changes:
---------------------

1) Search:
- aspiration window (search_full_root())
- razoring turned off (full_search())
- improved futility pruning (full_search())
- improved history pruning (full_search())
- improved LMR (history) (full_search())
- egbb hashing (idea by Teemu Pudas) (full_search())
- root LMR (full_root())
- improved extensions (full_new_depth())
- quiescence checks bugfix (full_quiescence())
- 5 killer moves (sort.h, sort.cpp)
- passer evaluation before LazyEval (eval())

2) Evaluation
- evaluation of weak squares (pawn_comp_info())
- improved pawn PST (pst_init())
- improved evaluation of knight mobility (eval_piece())
- improved evaluation of knight outpost (eval_piece())
- queen mobility (restored) (eval_piece())
- Lower multipliers in drawish positions (eval_draw(), material_comp_info())

3) Features
- automatical CPU number detection (init_threads())
- multiPV bugfix (idea by Norman Schmidt) (search.cpp, search_full.cpp)
- egbb bugfix (idea by Teemu Pudas) (full_search(), probe.h)

EGBB support:
------------
32-bit egbb dll should be named egbbdll.dll
64-bit egbb dll should be named egbb64dll.dll

Hashing:
-------
Note that hash size is critical in multi-CPU mode.
If hash is too small Grapefruit's performance on octal
will be slightly better than on a single core (NPS don't reflect actual speed).

Versions:
--------
Grapefruit 1.0 supports up to 8 threads.
Grapefruit 1.0 32cpu supports up to 32 threads.

Maybe it just explains the consolidated changes since Toga 1.4beta5c, and the latest was just the egbb bugfix (etc.) by Teemu for this version.

Have you compared Teemu's last fixed GrapeFruit1.0beta_y at R*bkaforum? This version was based on that one.

Thanks.

Denis

[ernest]

I agree, Jouni.

I am in the process of testing all these Toga-like versions (Grapefruit, Cyclone) against the latest Thinker (5.4 Inert).
They are just not as strong...

However, Tord's Glaurung 2.2 and its likeness Stockfish 1.2 opt. almost match Thinker, but I get incidents during some games (at least in the Fritz 11 GUI).

Have you experimented with Thinker, Jouni?

[Graham Banks]

I agree, Jouni.

I am in the process of testing all these Toga-like versions (Grapefruit, Cyclone) against the latest Thinker (5.4 Inert).
They are just not as strong...

However, Tord's Glaurung 2.2 and its likeness Stockfish 1.2 opt. almost match Thinker, but I get incidents during some games (at least in the Fritz 11 GUI).

Have you experimented with Thinker, Jouni?

Hi Jouni,

I've been having incidents in some games with these engines in the Fritz 11 GUI also, especially on my AMD64x2.
What incidents are you experiencing?

Regards, Graham.

[ernest]

Hi Graham,
Robert de Niro would say: "You're talking to me?"

or are you "talking to Jouni?"

In Thinker-Stockfish games, I had several Stockfish stops, in Thinker- Glaurung2.2 games, I had some Glaurung blunders...

[Graham Banks]

Hi Graham,
Robert de Niro would say: "You're talking to me?"

or are you "talking to Jouni?"

In Thinker-Stockfish games, I had several Stockfish stops, in Thinker- Glaurung2.2 games, I had some Glaurung blunders...

Ah yes - I should have been addressing you. Sorry Ernest.

I'm finding that Stockfish, Glaurung and Cyclone sometimes run slow all of a sudden, with the nps going right down and the search depth suffering as a result. I've had to replay countless games.
Sometimes Cyclone loses on time on the final move before the time control.
All this only happens on this computer, which is the only one of mine that is 64-bit.

[kranium]

Hi Graham,
Robert de Niro would say: "You're talking to me?"

or are you "talking to Jouni?"

In Thinker-Stockfish games, I had several Stockfish stops, in Thinker- Glaurung2.2 games, I had some Glaurung blunders...

Ah yes - I should have been addressing you. Sorry Ernest.

I'm finding that Stockfish, Glaurung and Cyclone sometimes run slow all of a sudden, with the nps going right down and the search depth suffering as a result. I've had to replay countless games.
Sometimes Cyclone loses on time on the final move before the time control.
All this only happens on this computer, which is the only one of mine that is 64-bit.

hi Graham,

if there anything i can do to help identify or resolve this issue...

you mentioned it only happens on your 64 bit computer...?
please give me as much info as you can and i will take a look.

(although the move time allocation code is no different than 141se and grapefruit...)
you never know ...often the smallest code change can have a domino effect, fix one thing, break something else.

Norm

[Graham Banks]

I'm finding that Stockfish, Glaurung and Cyclone sometimes run slow all of a sudden, with the nps going right down and the search depth suffering as a result. I've had to replay countless games.
Sometimes Cyclone loses on time on the final move before the time control.
All this only happens on this computer, which is the only one of mine that is 64-bit.

hi Graham,

if there anything i can do to help identify or resolve this issue...

you mentioned it only happens on your 64 bit computer...?
please give me as much info as you can and i will take a look.

(although the move time allocation code is no different than 141se and grapefruit...)
you never know ...often the smallest code change can have a domino effect, fix one thing, break something else.

Norm

Is there an SSE2 compile perhaps or is it already compiled for SSE2?

Regards, Graham.

[kranium]

I'm finding that Stockfish, Glaurung and Cyclone sometimes run slow all of a sudden, with the nps going right down and the search depth suffering as a result. I've had to replay countless games.
Sometimes Cyclone loses on time on the final move before the time control.
All this only happens on this computer, which is the only one of mine that is 64-bit.

hi Graham,

if there anything i can do to help identify or resolve this issue...

you mentioned it only happens on your 64 bit computer...?
please give me as much info as you can and i will take a look.

(although the move time allocation code is no different than 141se and grapefruit...)
you never know ...often the smallest code change can have a domino effect, fix one thing, break something else.

Norm

Is there an SSE2 compile perhaps or is it already compiled for SSE2?

Regards, Graham.

the 3.4 distribution (from cyclonechess.com) is highly optimized, but does not include any processor specific options...

if you like i would be happy to create it, and post it for download...
(consider it done)

[kranium]

it's cooking as we speak:

inlcudes: SSSE3, SSE3, SSE2, and SSE instructions...(optimized for Intel processors, like the following):
Intel® Core™2 Quad processors
Quad-Core Intel® Xeon® processors
Intel® Xeon® processors based on the Intel® Core™ microarchitecture
Intel® Core™2 Duo processors
Intel® Core™2 Extreme processors
Other Intel processors based on the Intel® Core™ microarchitecture

should be available for dowload soon...