Houdini 2

[veritas]

News about Houdini 2.0
[© M.C.Escher]

This page contains some preliminary information about the future Houdini 2.0 version. The information is provided without any warranty - dates and features can be changed at all times. The page will be updated as more information becomes available (current update was made on July 19, 2011).

Besides some Elo strength improvement and a number of minor bug fixes, Houdini 2.0 will focus on features that improve the user experience, more specifically for position analysis.
Another development axis is the support for servers with high number of cores (16 and up) that will become more readily available in the near future.

Release Date
The Houdini 2.0 release is scheduled sometime between September and December 2011.
Strength Limit feature
Houdini 2.0 will have a strength limit feature - from 0 (beginner) to 100 (full strength).

The strength is limited through a combination of techniques: limiting the number of positions searched, purposely picking a non-optimal move, and ignoring table bases.

Everyone should now be able to beat Houdini, we're sure happy we can now .
Improved Analysis support: Save Hash to File, Load Hash from File, Never Clear Hash
The complete hash table can be saved to a disk file and reloaded later for continuing the analysis.

Houdini 2.0 will make better use of the data in the hash table to restart instantly the analysis at the point where it was previously interrupted.

The goal is to be able to interrupt the analysis anytime and save the hash table in a disk file. At a later date you can reload the hash table file in memory and continue the analysis as if no interruption had happened.

A new "Never Clear Hash" option will keep the hash data in memory even when the position changes or when starting a new game.
Chess960 (Fischer Random Chess) Support
Houdini 2.0 will support Chess960 games.
Mate Search Limit
To speed up solving deep mates, you can fix a limit on the search depth during the mate search.
NUMA-aware version for high-end, multi-socket servers
For advanced servers with many cores on multiple sockets (usually with so-called NUMA architecture), a special "NUMA-aware" Houdini version will be available. It will significantly improve the scaling of Houdini when more than 8 cores are available, and therefore prepare for the next generation of CPUs with 16 to 32 cores.










http://www.cruxis.com/chess/houdini_news.htm

ppfff !!!!


btw

a little winged "bird" tells me NUMA-aware version(Beta) has a minimum 15% speed increase over 1.5a

[Jeremy Bernstein]

Wow. Some great, useful features for analysis. I'm glad to see that Robert is working on analysis/human-oriented features, and not just raw computer-computer optimizations.

jb

News about Houdini 2.0
[© M.C.Escher]

This page contains some preliminary information about the future Houdini 2.0 version. The information is provided without any warranty - dates and features can be changed at all times. The page will be updated as more information becomes available (current update was made on July 19, 2011).

Besides some Elo strength improvement and a number of minor bug fixes, Houdini 2.0 will focus on features that improve the user experience, more specifically for position analysis.
Another development axis is the support for servers with high number of cores (16 and up) that will become more readily available in the near future.

Release Date
The Houdini 2.0 release is scheduled sometime between September and December 2011.
Strength Limit feature
Houdini 2.0 will have a strength limit feature - from 0 (beginner) to 100 (full strength).

The strength is limited through a combination of techniques: limiting the number of positions searched, purposely picking a non-optimal move, and ignoring table bases.

Everyone should now be able to beat Houdini, we're sure happy we can now .
Improved Analysis support: Save Hash to File, Load Hash from File, Never Clear Hash
The complete hash table can be saved to a disk file and reloaded later for continuing the analysis.

Houdini 2.0 will make better use of the data in the hash table to restart instantly the analysis at the point where it was previously interrupted.

The goal is to be able to interrupt the analysis anytime and save the hash table in a disk file. At a later date you can reload the hash table file in memory and continue the analysis as if no interruption had happened.

A new "Never Clear Hash" option will keep the hash data in memory even when the position changes or when starting a new game.
Chess960 (Fischer Random Chess) Support
Houdini 2.0 will support Chess960 games.
Mate Search Limit
To speed up solving deep mates, you can fix a limit on the search depth during the mate search.
NUMA-aware version for high-end, multi-socket servers
For advanced servers with many cores on multiple sockets (usually with so-called NUMA architecture), a special "NUMA-aware" Houdini version will be available. It will significantly improve the scaling of Houdini when more than 8 cores are available, and therefore prepare for the next generation of CPUs with 16 to 32 cores.










http://www.cruxis.com/chess/houdini_news.htm

ppfff !!!!


btw

a little winged "bird" tells me NUMA-aware version(Beta) has a minimum 15% speed increase over 1.5a

[veritas]

my friends and i find what was not written slightly more interesting than was

ie he neglected / committed to mention the most widely asked question which has never been as he posts "will Houdini 2 be stronger re elo ?" but "Will Houdini 2 (by whatever name) be commercial ?"

On that note and being associated with certain "Beta testers" who have tested the NUMA architecture compiles i can see that tieing in quite nicely (for some ) with ChessBases supposed "New Revolution in computer chess " to be released around the same time frame as this engine , new cluster rentals feautering NUMA-aware engines in playchess @ a price

[Jeremy Bernstein]

my friends and i find what was not written slightly more interesting than was

ie he neglected / committed to mention the most widely asked question which has never been as he posts "will Houdini 2 be stronger re elo ?" but "Will Houdini 2 (by whatever name) be commercial ?"

On that note and being associated with certain "Beta testers" who have tested the NUMA architecture compiles i can see that tieing in quite nicely (for some ) with ChessBases supposed "New Revolution in computer chess " to be released around the same time frame as this engine , new cluster rentals feautering NUMA-aware engines in playchess @ a price

Possibly he hasn't decided yet. Possibly he knows that revealing pricing and/or cooperation with CB will ramp up attention to certain unorthodoxies wrt Houdini's origins, and potentially throw a monkey wrench into the works. Possibly it's all for free. It's fairly normal for developers to keep release details private until the last minute, so I don't find this so unusual.

Considering the amount of work that Houdart is putting into Houdini, it would honestly surprise me if he didn't start charging for it. Since the IPPOLIT code is public domain, and since there's still no consensus on where that code comes from, there's nothing objectively illegal about it, either, but I maintain that it's pretty lame. OTOH, he's brought the IPPOLIT code base much further than anyone else, both strength and feature-wise. Ah, ethics...

Jeremy

[BB+]

Considering the amount of work that Houdart is putting into Houdini, it would honestly surprise me if he didn't start charging for it. Since the IPPOLIT code is public domain, and since there's still no consensus on where that code comes from, there's nothing objectively illegal about it, either, but I maintain that it's pretty lame. OTOH, he's brought the IPPOLIT code base much further than anyone else, both strength and feature-wise. Ah, ethics..

There is some question whether an IPPOLIT re-packager could be a "vicarious" infringer of copyright [somewhat like when buying cut-rate goods in a circumstance that should warn one that they are likely stolen]. This is particularly true in places where RE is verboten.

[veritas]

Considering the amount of work that Houdart is putting into Houdini, it would honestly surprise me if he didn't start charging for it. Since the IPPOLIT code is public domain, and since there's still no consensus on where that code comes from, there's nothing objectively illegal about it, either, but I maintain that it's pretty lame. OTOH, he's brought the IPPOLIT code base much further than anyone else, both strength and feature-wise. Ah, ethics..

There is some question whether an IPPOLIT re-packager could be a "vicarious" infringer of copyright [somewhat like when buying cut-rate goods in a circumstance that should warn one that they are likely stolen]. This is particularly true in places where RE is verboten.

+1

but less we forget

Houdart stated publicly " He would post the source code once he had cleaned it up " so hes either a BARE FACED LIAR or incapable of removing all traces of the codes he SWEARS hes (JOKE ) " NEVER EVEN LOOKED AT"

[Uly]

The save/load hash feature as depicted for Houdini 2 seems very much inferior to other "learning" methods such as Shredder's position learning, Rybka 3 Persisntent Hash (without the bugs) or Private Critter, in where the user doesn't need to invoke a save command every time he unloads the engine and the hash isn't polluted with old entries that don't need to be there.

Why does one need to save the whole hash when just saving PV and internal PV entries suffices to get the desired behavior? Why would one need to have one file the size of given RAM for every game one is analyzing when a single file (of arbitrary size because the engine can be selective on what is being saved) for all games could be sufficient? Why would the user need to interact to save anything when the engine could do it automatically and in the moment it is needed? (I can imagine someone getting a power outage and losing all the unsaved analysis because the engine will not do it by itself).

I think Robert is going with the wrong approach for continuing the analysis after unloading the engine.

I'm still happy that he's willing to try to make Houdini more analysis friendly, I guess I can't ask for more if he's giving his best, I just find it unfortunate that he decided to do "learning" on a clumsy manner.

[TPJR]

The save/load hash feature as depicted for Houdini 2 seems very much inferior to other "learning" methods such as Shredder's position learning, Rybka 3 Persisntent Hash (without the bugs) or Private Critter, in where the user doesn't need to invoke a save command every time he unloads the engine and the hash isn't polluted with old entries that don't need to be there.

I think Robert has read your post:

"The "Position Learning" mode will automatically save analysis results in a learning database that will be reused in future analysis. This is a more convenient, but slightly less powerful method of using previous analysis results than the manual Save Hash/Load Hash operations of the previous point."

[Uly]

Awesome!

This is a more convenient, but slightly less powerful

It's slower as the engine has to research what hasn't been stored, but for analysis what is important is being able to propagate the right scores of the moves anywhere.

It seems Houdini IS getting learning! Now I'm very interested in the new release

[Prima]

Awesome!

This is a more convenient, but slightly less powerful

It's slower as the engine has to research what hasn't been stored, but for analysis what is important is being able to propagate the right scores of the moves anywhere.

It seems Houdini IS getting learning! Now I'm very interested in the new release

Nice. Thanks Uly for bringing this to RH's attention. Thanks Mr. Houdart for listening and willing to try this route. I'm confident Houdini will be much better at analysis than we can imagine. Not that I'm the better judge....