Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / Houdini 6
- - By Boombaard (**) Date 2017-09-25 18:46
How does the new Houdini 6 fares?
Is it stronger than Stockfish, Komodo?
Parent - - By Labyrinth (*****) Date 2017-09-25 23:25
Quite well. Seems to be stronger than both Komodo and Stockfish 8. Not as certain about SF development version/asmFish and such, but on some tests I think it still pulled ahead. So it's probably the #1 engine at the moment.
Parent - - By RFK (Gold) Date 2017-09-26 01:01
Labyrinth, I have to wait until next week before I can buy Houdini 6. If you have some time see how it fairs playing against SugaR XPrO. 1.2 64 BM12. It appears to be stronger than Stockfish 8.
Parent - - By shrapnel (***) Date 2017-09-26 18:01
It appears to be stronger than Stockfish 8
Stockfish 8 is a Relic and completely unimportant.
Performance against latest Stockfish Development Version is what counts.
Parent - - By RFK (Gold) Date 2017-09-26 18:09 Edited 2017-09-26 18:17
I did some limited testing on the last development version and is on par with Stockfish 8. But I have to preface that in saying I didn't waste my time going too far into any extensive game testing. It was limited.

CumnorChessClub(Kevin D Plant) was in fact correct in his post

SF has flat-lined in their progressed. I had to find that out for myself.
Parent - - By Labyrinth (*****) Date 2017-09-26 21:42
Looking through various tests on talkchess seems like H6 is losing to latest SF dev/asmFish at some time controls on some hardware. They're always really close though, so it could be a number of games issue as well. Either way, they're both really goddamn strong.
Parent - - By RFK (Gold) Date 2017-09-26 22:23
I downloaded Something that claimed to be asmfishW. But there is no version notation connected with it?
Parent - - By Peter Grayson (****) Date 2017-09-27 01:07 Edited 2017-09-27 01:09

> I downloaded Something that claimed to be asmfishW. But there is no version notation connected with it?

The Windows asmFishW Engines can be downloaded from here ...

The 2017-05-22 engine has performed best in my tests up to last release.

Parent - By RFK (Gold) Date 2017-09-27 01:33
Thanks I got it. Appreciate you help!
Parent - - By Peter Grayson (****) Date 2017-09-27 01:05

> H6 is losing to latest SF dev/asmFish at some time controls on some hardware

May be worth checking out the balance of openings too. Stockfish engines generally get better results with certain King's pawn openings, particularly the Sicilian and Spanish from my opening set but Houdini performs better in non King's Pawn openings. I am finding the same with Houdini 6.01 Pro in current Gauntlet when Stockfish 8 is giving it a tough time but in contrast the performance against Komodo has been eye opening so far.

Another adverse influence will be making the same mistake as I did by having 6 man Nalimov available and only 5 man Syzygy. Despite the "6" engine issue with Nalimov being corrected in "6.01", the slow down in "6.01" when Nalimov are accessed in the absence of appropriate Syzygy bases can be quite dramatic. Worst so far noted using six cores per engine was a drop from 13,500 kN/s to as low as 55 kN/s and plydepth 15 ply lower. That will not be seen in the standard engine because it will not use Nalimov. The problem is the Chessbase GUI sends the engine Nalimov paths despite being cleared from the GUI entry. A work around is to use empty dummy paths that overwrite the originals and that works but too late to implement in current Gauntlet.

Parent - By RFK (Gold) Date 2017-09-27 06:51
I just started testing -I'm not getting too serious about this. SugaR XPrO 1.2 vs AsmFish 2017 etc. They are coming out drawn that goes for Stockfish 8 as well. It is only on low end hardware that SugaR XPrO 1.2 beats the crap out  asmfish so far in my albeit limited tests. But when I move into 6900k the both of them even far! That could change by morning.

I'm only playing a 25 minute games. I'm not going to tie up my computer for a great length of time.
Parent - - By RFK (Gold) Date 2017-09-27 21:41 Edited 2017-09-27 21:50
So far 12-25 minute games SugaR XPrO 1.2 64 BM12 vs asmfish_2017-08-02_bmi2 10 draws one win for SugaR.
I'm trying for experimental purposes a French  Defense Power book. I'll try an number of exclusive opening books about 25 games each. See what happens.

2048 hash
Ponder off
Hyper threading off.
Turbo is off.
i7-6900K 3.20GHz
32 GB ram
Parent - - By Peter Grayson (****) Date 2017-09-27 21:48
Would be interesting to know what kN/s you get from the start position.

Parent - By RFK (Gold) Date 2017-09-27 21:54 Edited 2017-09-27 22:02
I've eight minutes left in this one game. Does sedat still have his area test setup? I have the sedat Stockfish Area benchmark from sedat's site. I wonder if I can use it to determine kN/s from the opening position. I don't see why not!

The 12th game doesn't appear to be looking like a possible drawn game so I'll let play out.
Parent - By RFK (Gold) Date 2017-09-27 22:11 Edited 2017-09-28 07:32
My mobo does a lot of the work for me in terms of balancing any overclocking. I just cranked once and was impressed with the increase very minor voltage just about imperceptible. Temps were silly mild. Height of summer in my apartment at 80 degrees. and the temps stayed at 43c across the board. What was cranking a "bit high" not much -but enough for me to see an increase was the wattage. My fans were whizzing -I mean they were Whizzing. Mobo remain 32c. kN/s increases perceptibly! But the fans were going nuts. Fans went from silent to ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZz good thing they are ball bearing fans.

Addendum: I attribute the mild temps to a really great water cooling block from EK-Supremacy EVO Elite Edition - Intel 2011-3! Excellent!
Parent - By RFK (Gold) Date 2017-09-27 22:50 Edited 2017-09-27 22:58
I think this may be what you are looking for -I did a couple tests -I might have been a little off on my previous estimation. I gave it one minute -I think that is correct -isn't it?
Sedat's was automatic cutoff. I think if I overclocked I could easily get 22-23000 kN/s.
Parent - By RFK (Gold) Date 2017-09-27 23:02
Good thing I didn't interrupt that last game! :wink:
Parent - - By RFK (Gold) Date 2017-09-27 23:09
What do you guy's get around 40,000 kN/s on a duel Xeon server setup?
Parent - - By jpqy (**) Date 2017-09-28 05:56
You can look here a new benchmark from Ipman using asmFish 2017-05-22 with last new CPU's
Just dubble click asmFish and type bench 1024 16 26 (enter) you just change 16 as your cores/threads you want to use!

Parent - By RFK (Gold) Date 2017-09-28 07:21
Thanks for the link!
Parent - - By RFK (Gold) Date 2017-09-29 18:54 Edited 2017-09-29 19:16
Okay! I want your take on this. We keep coming back to this issue time and time again! The use of  IntelĀ® Hyper-Threading Technology in computer chess -whether in engine vs engines matches or in analysis -does hyper threading help or hinder?

Addendum: I noted that Robert Houdart used hyper threading  in his analysis of a position. Given his high end hardware not too much is going to be missed in search  - hyper threading , I would imagine cuts down his search time?! "

Dual Xeon E5-2698v4 @ 2.3 GHz, 80 threads, 32768 MB of Large Page hash, 6-men Syzygy.
74 MN/s on a board full of pieces, after 20 hours about 5.4 trillion nodes.

This is just a question. I'm not sure one could miss anything of significance running at this level of search this fast- it's  mind boggling at 5.4 Trillion nodes!
Parent - By Peter Grayson (****) Date 2017-09-29 20:54

>Okay! I want your take on this. We keep coming back to this issue time and time again! The use of  IntelĀ® Hyper-Threading Technology in computer chess -whether in engine vs engines matches or in analysis -does hyper threading help or hinder?

With asmFish I get around the 30,000 kN/s mark using a dual Xeon E5-2687W 2 x 8 cores with HT switched off.

When I looked into the "HT On or Off" debate after purchasing the machine I ran some position solving tests. It was my view that the best measure was "time to solve" as the reference measured variable and not kN/s that is really something of a nothing measurement given historic engine misreporting. However a single "time to solve" measurement for comparison when threads > 1 is insufficient because of the inherent variability of muli-threaded operation. Each measurement will give a different time. Therefore I decided to run each position 100 times and average the time out.

With the engines available at the time the results were inconclusive. i.e. there seemed no significant benefit because there was little if any measured improvement from the extra logical cores of hyper threading. However there was noticeable extra heat being generated by the machine confiming higher power usage for no benefit.

I actually revisited this a few weeks ago to look at the asmFish engine multithread efficiency and was quite shocked to find on my quad, when testing performance using the T2830 position 15, bm Kg3 compared to the single thread 55s solve time, using 4 cores the averaged out solving performance was worse than the single thread performance. It suggested something else was happening that could not be measured by previous tests. The development Stockfish code date stamp equivalent was tested with the same result.

Given slightly less than a minute solving time for single thread I thought an 8 minute cut off solving time would capture the high variability times using 4 cores. It was insufficient! In the first 40 tests the engine failed to solve 28 times in 8 minutes what the engine solved with a single core. This meant the MP perfromance was much worse than single core. I have not got back to this to try and understand what is going on with my attention now diverted with the flurry of updated engines.

Therefore based on my earlier tests I was unable to find any convincing evidence there is much, if anything to be gained from running with HT on. Others with later hardware may be able to show otherwise.

Parent - - By Stonehenge (***) Date 2017-09-29 22:53 Edited 2017-09-29 23:41
I've recently run 3 hyper-threading tests with Houdini 6:
- 12 threads vs 24 hyper-threads on 12 cores (Xeon)
- 20 threads vs 40 hyper-threads on 20 cores (Xeon)
- 40 threads vs 80 hyper-threads on 40 cores (dual Xeon)

In all 3 tests the hyper-threading adds about 10 Elo in strength.
The results are documented somewhere in this Talkchess thread:
Note that this is a relatively poor gain for a 30% node speed increase, doubling the number of threads generates some inefficiency (about 15%).

Some people in the Talkchess thread were claiming that there would be no advantage in using higher number of threads (a claim based on "Amdahl's law").
My results show that this is incorrect, and that there is benefit in using more threads, at least up to 80 threads with Houdini 6.
Parent - - By RFK (Gold) Date 2017-09-30 02:39
I'm not sure how much of an increased gain in elo would be generated on low end hardware such as an i7-6900K. I intend to buy the Houdini 6 Pro regardless whether I could get away with using the standard.
The level of inefficiency may be either the same percentile you cited?! It is hard to say without testing. Overclocking on low end may be the only way to go assuming one has a reasonably good setup and keeps within safe overclocking cpu limits. I've pushed two units for 5 yrs to their limit and black screened them. I'm retired now and am not so hard on things as I used to be.
Parent - - By APassionforCriminalJustice (***) Date 2017-09-30 03:17 Edited 2017-09-30 03:21
Why on earth are you calling a 6900k low end? You get over 17000 passmark score - and it's 8 cores. There is nothing low end about that.

From my understanding hyperthreading has improved with recent Intel architectures. I wouldn't feel comfortable using hyperthreading on say Peter's hardware because it is Sandy Bridge. But with say Haswell or Broadwell then why not? 30 percent increase in nodes will give you something at least.
Parent - By RFK (Gold) Date 2017-09-30 03:42
Well, you know, compared to what you guys are using, I'd have to consider it low end.

I don't mind doing some testing using hyperthreading. I might play with the over clocking. I get excellent temps on the Broadwell. Where as the Haswell chips get extremely hot fast and are hard to cool. The Sandy bridge weren't too bad.
Parent - - By Stonehenge (***) Date 2017-09-30 09:37
An i7-6900K is an excellent CPU.
Hyper-threading results will be similar to my hardware, you gain about 30% in total node speed but half of the gain is wasted because of the increased thread number.
Parent - By RFK (Gold) Date 2017-09-30 15:31
With that said, I think I will try carefully overclocking my machine incrementally and see where it takes me. All in all I think it  would give me a more balanced and efficient performance than if I just used hyper-threading. It just may take me a little time to work up to getting back into playing with the bios. I don't want to fry this one! :lol:
Parent - - By Mark Eldridge (****) Date 2017-10-03 12:34
Does H6 support Syzygy bases.
Parent - By Mark Eldridge (****) Date 2017-10-04 12:08
Parent - - By RFK (Gold) Date 2017-10-06 02:35
Hey! Robert? Is  Houdini programmed to run with all 32 pieces of white and black - or, do I have to imagine any one particular piece move in a game with an arrow? If so do I get to choose the color of the arrow?
Parent - - By darmar (**) Date 2017-10-06 10:06
It depends of your overclock. For full playing with 32 pieces you need 4GHz minimum. For using an arrow in your book must exists red  colored moves for both sides.
Parent - By RFK (Gold) Date 2017-10-06 15:15
That answer is as good as my question! :lol:
- - By Hamster (**) Date 2017-10-01 13:52
Did anyone compare the similarity of Houdini 6 with Stockfish 8 resp. the development builds? I wonder if this version is more similar than Houdini 5 was.
There used to be a relatively easy to use tool (similar.exe I think) for those similarity comparisons...
Parent - - By darmar (**) Date 2017-10-01 15:34
Houdini 6 is 99% Sf
Parent - By APassionforCriminalJustice (***) Date 2017-10-01 17:27
Lol is that so? Why then was Houdini stronger than Stockfish years ago? I still remember Houdini beating Stockfish in TCEC Superfinal. Houdart is just a super-talented developer.
Parent - By RFK (Gold) Date 2017-10-01 18:36 Edited 2017-10-01 18:39
darmar, as you well you know. Saying something is easy. It is in supporting what you say with substantial facts- substantive data that moves one away from just making   hypothetically premised opinioned statements- and moves us into fact based knowledge.

The U.S currently has a president that thrives on fake news.
Every prof will ask "where is your data?!"
Parent - - By RFK (Gold) Date 2017-10-01 19:06 Edited 2017-10-01 19:10

> Houdini 6 is 99% Sf

When it is you making the accusation -it isn't going to be Robert Houdart that has to prove anything,  one way or the other, that he's done anything wrong it is going to be incumbent upon you to do the work.

You are going to have to come up with undisputable  evidential documentation.

I've felt that one or two other developers have cheated along the way-no doubt about.

One I've even gone to bat for-if only because I thought the way the programmers in the programming community went after him was without moral principles.
In fact, most were probably guilty of the same charge they were accusing Vasik Rajlich of , but, yeah, sure, more than likely to some lesser degree. But, the fear of being found out! They wanted to  bury Vasik Rajlich as quickly as possible -even if it meant burying him " theoretically " alive - and so they did.

This is an impure game with a hell of a lot of politics involved -there is nothing new about that. Surprise! Surprise!
Parent - By InspectorGadget (*****) Date 2017-10-05 14:37

> When it is you making the accusation -it isn't going to be Robert Houdart that has to prove anything,  one way or the other, that he's done anything wrong it is going to be incumbent upon you to do the work.

I like this :lol::lol:
Parent - - By Labyrinth (*****) Date 2017-10-01 22:01
Parent - - By Peter Grayson (****) Date 2017-10-06 18:18
Got Talent Rating ...
Parent - By RFK (Gold) Date 2017-10-06 18:40
- - By nbeqo Date 2017-10-06 01:47

I am using Rybka chess program, Houdini 6.02 engine, and have my own Opening Book for White in a pgn file and for Black in pgn file.
Can someone please help me customise the existing opening book (Rybka or Houdini Book) by merging my pgn files and going forward also to keep adding moves to the opening Books
Or is there a Graphical interface that does just that, saves moves to the opening book?
I am pretty sure any GM would love such an interface for their opening preparation rather than investing in chess programs and engines.

Thanks in advance!
Parent - By darmar (**) Date 2017-10-06 10:08
What Rybka chess program? Rybka Aquarium with Houdini openbook or Ribka - Fritz GUI with ctg book?
Up Topic The Rybka Lounge / Computer Chess / Houdini 6

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill