How strong is the new Fire 1.5 xTreme?

[Frank Quisinsky]

Hi Norman,

SWCR Fire 2.0 x64 xTreme test started tomorrow to the same time with SF 2.1.1 JA PHQ-2. I am very interested now, let us see. Need around 8 days (two engines to the same time).

Thanks for your new version and the very good documentation about all the changes you do.

Best
Frank

PS: I start testing with the default parametes. I will wait a bit and later a second test with a good setting is possible. Wait here of a message from Dieter Eberle. Important for my rating list is to have at first a result from the default settings.

[Dr.Wael Deeb]

As you noticed,I've changed all the pieces values using a well tested formula plus changing the king's safety parameter,the mobility and the positional weights....
Dr.D

I don't suppose you're willing to share your piece value formula? I'm extremely interested in piece value tuning, and have tested every piece value set i could find.

I noticed your king's safety parameter, mobility weight positional weight are almost identical to parameters i already have, from a post on another forum. That poster must have been you,right?

Regards Dave

Not quite David....I was just inspired by these parameters and wanted to combine them with my pieces value formula which is already showed great results with a FireBird personality....
Dr.D

[Dr.Wael Deeb]

just a quick note....

i'm releasing Fire 2.0 xTreme as we speak...
i certainly recommend that users now use this latest version.
any fire.cfg file or saved UCI option form a previous version should be 100% compatible with 2.0

there likely no big ELO increase, but several issues have been resolved.
largest of which was the occasional Playchess.com crashing.

major changes:
now supports up to 64 cores/threads
occasional crash fixed
code improvements, bitwise optimizations
SEE piece/capture values are no longer constant entities
fire.cfg incorrect default value for Prune_Pawn fixed: Prune_pawn = 160
Fire now recognizes whether or not the system supports hardware popcount
and/or large pages, and self-configures itself accordingly
some code cleanup, formatting, and re-organization
more complete Log implementation
UCI clear hash button added
etc.

at the moment this is about as good as I can get it...

i have had a tremendous response and I want to thank all concerned...
many have emailed me with settings/parameters that they claim now equal Houdini,
especially when using a large powerful system.
that is of course very gratifying, as it was my goal during the creation of these xTreme configuration
releases.

for the future, i do hope the configurations are shared...
and eventually brought over to any new version of IvanHoe that may be created.

Best Regards-
Norm

Thanks Norman.....will continue testing with the new version using my Scylla parameters....
Dr.D

[kranium]

Thanks!

But: Multi-pv is broken (at least in the 32-bit version, in XP with P4 cpu)

In Fritz(12), neither changing via interface (+/-) nor in the UCI options aka engine preferences has an effect on the number of analysis variations.

(Triplebase access works for me with Fire 2.0, in Fritz.)

In Arena 2.0.1, the multi-pv selection in Arena's engine context menu is greyed out. So I assume the interface doesn't get information that the engine is multi-pv capable. I can change the multi-pv number in the engine options, and then Fire 2.0 does multi-pv. But the --- line between the iterations is missing.

(I am not using a fire.cfg)

Fire 1.5 works ok here regarding multi-pv, basically (except for a Fritz problem which clearly is not Fire specific).

Otherwise, I'm looking forward to great results. The gap to Houdini 1.5a was not very big anymore...

damn
fix one thing break another...
(15 beta testers and nether I nor any of them caught this!)
thanks Mike for the heads up ...
guess i''ll have to issue a maintenance release...
2.01
coming soon tho a web site near you...

[Frank Quisinsky]

Hi Wael,

remember:
Your Fire 1.1 settings are 10 ELO weaker as Fire 1.31 under SWCR conditions. 4 of this 10 ELO comes from a very high draw quote. 5% more as Fire 1.31. Tested after 1.000 SWCR games. Calculated with Bayesian, draws are here more important compare to ELOstat.

But ...
More passiv = better in blitz.
Possible that your setting with extrem short time controls are better as Fire 1.31. Absolutly possible!

Best
Frank

I think the results can be found in TalkChess tournament selection. The games I deleted later from my rating list.

[kranium]

Hi Norman,

SWCR Fire 2.0 x64 xTreme test started tomorrow to the same time with SF 2.1.1 JA PHQ-2. I am very interested now, let us see. Need around 8 days (two engines to the same time).

Thanks for your new version and the very good documentation about all the changes you do.

Best
Frank

PS: I start testing with the default parametes. I will wait a bit and later a second test with a good setting is possible. Wait here of a message from Dieter Eberle. Important for my rating list is to have at first a result from the default settings.

Hi Frank-
like I said, there is likely no great ELO increase...
(a small one is possible because of the code optimizations)

for ex:
here, on my test system, Fire 2.0 consistently wins against IvanHoe 47c (w/ short TCs)
~ 53%/47%, but it remains to be seen if this holds up...

i do have some very interesting configs from some of the most active playchess.com users though, one of which changes Fire into an aggressive and tactical monster...Serpent from Nolan Denson who posted earlier in this thread. i haven't tested it much, but he and others have played many playchess games with great success. The few games I witnessed surprised me...the way Fire played was completely different than anything i had seen in the thousands of games i've watched previously...(really a somewhat of a revelation)

(personally...i don't like parameter tuning, and have not done it with Fire...find it too tedious, (although LittleBlitzer should be a great help) it's a job begging to be done.

i believe there's great potential for improvement for someone with the ability and patience to do it, and i think it's interesting for the users to play/experiment with.

Many thanks and best regards to you-
Norm

[Frank Quisinsky]

Hi Norman,

thanks for your explanation. I think Dieter Eberle will give me later the information which setting I have to test. No contacts to Nolan and I am not a Fire expert. I think Dieter will find out it and I test it

For the moment the default settings I should test.

Yes, with all the settings and the playing style the potential I saw too. More aggressiveness could be give Fire a higher rating, think so!

Best and thanks
Frank

[kranium]

Yes, with all the settings and the playing style the potential I saw too. More aggressiveness could be give Fire a higher rating, think so!

yes, you mentioned this previously.
your insight and advice is spot on i think.

[kranium]

Thanks!

But: Multi-pv is broken (at least in the 32-bit version, in XP with P4 cpu)

In Fritz(12), neither changing via interface (+/-) nor in the UCI options aka engine preferences has an effect on the number of analysis variations.

(Triplebase access works for me with Fire 2.0, in Fritz.)

In Arena 2.0.1, the multi-pv selection in Arena's engine context menu is greyed out. So I assume the interface doesn't get information that the engine is multi-pv capable. I can change the multi-pv number in the engine options, and then Fire 2.0 does multi-pv. But the --- line between the iterations is missing.

(I am not using a fire.cfg)

Fire 1.5 works ok here regarding multi-pv, basically (except for a Fritz problem which clearly is not Fire specific).

Otherwise, I'm looking forward to great results. The gap to Houdini 1.5a was not very big anymore...

damn
fix one thing break another...
(15 beta testers and nether I nor any of them caught this!)
thanks Mike for the heads up ...
guess i''ll have to issue a maintenance release...
2.01
coming soon tho a web site near you...

MultiPV has now been fixed, will be releasing maintenance version 2.01 tonight
Thanks Mike for bringing this to my attention.
Apologize for any inconvenience

[Mike S.]

Thanks again! A very fast bugfix which should be awarded with +30 bonus Elo on the rating lists.