I Am Running Strelka 5.6 tests- Promise Accurate Results!

[geots]

As we speak, I am testing Strelka 5.6., as there was a copy from Jury in my email. He knows my testing and that he can expect good results. He knows I am and have been one of his most vocal supporters. But most important- he knows he can trust me with his life- so certainly his version is as safe in my hands as it is in his. I will not disappoint- my credibilty is the most important thing I possess.

Results will certainly be posted here for anyone interested.



Best,

george

[lucasart]

As we speak, I am testing Strelka 5.6., as there was a copy from Jury in my email. He knows my testing and that he can expect good results. He knows I am and have been one of his most vocal supporters. But most important- he knows he can trust me with his life- so certainly his version is as safe in my hands as it is in his. I will not disappoint- my credibilty is the most important thing I possess.

I'm not sure I understand what you're saying. Do you mean to say that Yury Osipov emailed you Strelka 5.6, and wants the engine to remain private?

If it's a release candiate, I can totally understand his point of view. Very often, when I released a version people started testing it, and then I found a bug, so testers were annoyed to see that a bugfix version followed a couple of days after. So it makes a lot of sense to give a release candiadate to selected beta testers, who will report bugs and results, before going for mass distribution.

But once stabilized, and tested, I do hope that Yury Osipov decides to make Strelka 5.6 publicly available. Really, what's the point of making a private engine ? So you write a program, keep it secret, and no one can use it... A program that cannot be uses is... erm... useless, no ?

PS: Can you stop writing everything in bold ? It's kind of annoying to read bold text.

[geots]

As we speak, I am testing Strelka 5.6., as there was a copy from Jury in my email. He knows my testing and that he can expect good results. He knows I am and have been one of his most vocal supporters. But most important- he knows he can trust me with his life- so certainly his version is as safe in my hands as it is in his. I will not disappoint- my credibilty is the most important thing I possess.

Results will certainly be posted here for anyone interested.



Best,

george

I will give Lucas a break here. In a manner of speaking. To give you an idea of the Strelka 5.6 testing, here are the beginning matches:


50-game matches, 1 CPU

Against:

Houdini 3 x64
Critter 1.6a 64-bit
IvanHoe 50kQ x64 SSE 4.2
Ivanhoe 46h
Komodo 5 x64
Fire 2.2b xTreme GH x64

The above 6 matches will all be run at 4'+2", simply because Strelka 5.5 has a 50 game match at that time control ag. each of the above. And to get a good comparison using 5.6, the controls should be the same.

When completed, we shall compare Strelka 5.5's results to Strelka 5.6



After the completion of the above, these will follow. Again, 50-game matches. But they likely may be run at 40/4 repeating, as I have no matches ag. Strelka 5.5 with any of them to compare to 5.6. Then again- possibly 4'+2". At any rate, they are:

Stockfish 2.3.1 pcnt x64
Rybka 4.1 x64
Houdini 1.5a x64
Komodo 4 x64
DeepSaros 3.3 x64 (exact same as 3.3b)
Bouquet 1.6 pop64



Upon completion of these, we will try some 40/20 repeating and/or 40/40 repeating. When I am thru- Strelka 5.6 will be an extremely well-tested version!



george

[geots]

It is only one change, but a fairly decisive one. Instead of the 50-game matches I stated would be run, the change has been made and they are now all 100-game matches.

I notices that Lucas complains about the bold text, which is so much easier on me. The light print doesn't help my eyes. And mine is the only print I can change. But right now I will meet him halfway. Hence "half-and-half" here.

By the way, good progress is being made. The first 4 matches are near the halfway mark now.

And I mentioned that Yuri was nice enough to grant me permission to post the results for any or all who may be interested.



Nite-

george

[Kyodai]

Thanks George! These are extremly interesting matches!

As for bold or not - choose the text best suited for your own eyes.

(No problem here...)

[geots]

Thanks George! These are extremly interesting matches!

As for bold or not - choose the text best suited for your own eyes.

(No problem here...)

Thank you Sune for the very nice reply. It is a good time for me to say that a little later this evening I will be posting the results from the first 4 matches. I know all about + and - error margins- in this case I am not interested. What you see is not only "what you get", but also "what you will continue to get" if you replayed the matches 100 times. Some things took place in these matches that I would have said were just not possible. No one has run the Ivanhoes anymore than I have. In fact I spent 2 or 3 months setting up a rating list of the top 10 64-bit versions, with the help of Peter Osterlund. Ivanhoe 46h was by far the strongest- tho with the recent release of 50kQ- that might be questionable.

At any rate, I cannot count the matches I have run with Izak's (PP) 46h. I have not run it ag. Houdini 3- so there I cannot say. But Houdini 2 could never come close to what I have seen take place here. NOT even close. NO ONE pushes 46h around like this.

Maybe what is the most astounding, Yuri has come up with a unique and original way to handle openings. Never seen it before, and I am surprised now that no one else has ever thought of or considered it. It in itself is likely worth more than a dozen elo points per game.

Please stay tuned Sune. This is worth waiting for. Trust me.



Best,

george

[geots]

As I said to Sune in the previous thread, Yuri has found an ingenious handling of openings- maybe it is the "new opening theory" for engines.

What Strelka 5.6 has done in these matches- I still pick Ivanhoe 46h as the most astounding result- when you consider that the Ivanhoes were basically born and bred for blitz chess. No one has EVER pushed 46h around like this. NOT EVEN CLOSE.

I am not sure of the importance attached to these comparisons, but it is really all I have from Strelka 5.5 and these 4 engines. Maybe there is no importance. But I wouldn't take Strelka 5.6 lightly- I'm just sayin.......... At any rate:

Code: Select all

Strelka 5.5 - Fire 2.2b xTreme GH x64   (26.5-23.5)
              Komodo 5 64-bit   (25.5-24.5)                   
              Ivanhoe 46h x64   (28.5-21.5)
              IvanHoe 50kQ x64 SSE4.2   (26.5-23.5)

Inspiron 620 Intel i5-4 True Cores
Fritz 11 gui/Fritz 13 gui
1CPU/64bit
128MB hash
Bases=NONE
Ponder_Learning=OFF
Perfect 2012b.ctg w/12-move limit
4'+2"
Match=100 games

Code: Select all

SP8-Core-i5, Blitz:4'+2"  2/13/2013  

                             
Strelka 5.6 x64           +63    +31/-13/=56   59.00%   59.0/100
Fire 2.2b xTreme GH x64   -63    +13/-31/=56   41.00%   41.0/100

Code: Select all

SP8-Core-i5, Blitz:4'+2"  2/13/2013   

                     
Strelka 5.6 x64   +78    +38/-16/=46   61.00%   61.0/100
Komodo 5 64-bit   -78    +16/-38/=46   39.00%   39.0/100

XPS 8500 Intel i5-4 True Cores
Fritz 11 gui/Fritz 13 gui
1CPU/64bit
128MB hash
Bases=NONE
Ponder_Learning=OFF
Perfect 2012b.ctg w/12-move limit
4'+2"
Match=100 games

Code: Select all

XPS-Core-i5, Blitz:4'+2"  2/13/2013


Strelka 5.6 x64   +92    +31/-5/=64   63.00%   63.0/100
Ivanhoe 46h x64   -92    +5/-31/=64   37.00%   37.0/100

Code: Select all

XPS-Core-i5, Blitz:4'+2"  2/13/2013  

                             
Strelka 5.6 x64           +42    +27/-15/=58   56.00%   56.0/100
IvanHoe 50kQ x64 SSE4.2   -42    +15/-27/=58   44.00%   44.0/100

4 matches: +63, +78, +92, +42 - And except for the "ringer, ie basically unknown" IvanHoe 50kQ, it was just complete strength and devastation.




It may be a slight change. But the next 4 matches beginning shortly at the same above controls will be against:

Rybka 4.1 x64
Stockfish 2.3.1 x64
Critter 1.6a 64-bit
Houdini 2.0c x64 (Tho I have results where Strelka 5.5 beat Houdini 2.0c in a 50-game match earlier by the score of 26.5-23.5- it is just really, really hard for me to see Strelka 5.6 being able to duplicate that feat. There had to be luck involved in that match- I would think)



These 4 matches will then be followed by, at the same blitz controls:

Bouquet 1.6 pop64
DeepSaros 3.3 x64
Komodo 4 x64
***Houdini 3.0 x64- (If, and that is a huge "if", Strelka 5.6 were to get by Houdini 2.0c- certainly all eyes would be on the this "Super Match"!)



After the above "2, 4-match sessions" left- I have run out of engines to give a shot at the gold. Unless someone has ideas of a few I forgot about- we shall have to look at "same opponents, different time controls". So far, Strelka 5.6 has really not even come close to being tested.




Enjoy- time for me to get the next 4 matches started,

george

[Modern Times]

Code: Select all

SP8-Core-i5, Blitz:4'+2"  2/13/2013   

                     
Strelka 5.6 x64   +78    +38/-16/=46   61.00%   61.0/100
Komodo 5 64-bit   -78    +16/-38/=46   39.00%   39.0/100

That is a very good result.

[Dan Astrachan]

Georges, two questions:

- Does Strelka 5.6 give "weird" evaluations like 5.5 did?
- Do you have any information as to a possible release date?

[geots]

Georges, two questions:

- Does Strelka 5.6 give "weird" evaluations like 5.5 did?
- Do you have any information as to a possible release date?

Hi Dan-

I had all 4 matches running at once, and did not get a chance to watch a lot of any of the games. And a bit over half of each one finished while I was sleeping. I did as always check the PGNs to make sure there were no crashes and all wins were good ones with no time losses.

I use the same Sedat generic book for all- with a 12 move limit. But Yuri has come up with an extremely novel way of playing the opening. I have never seen anything like it ever done before. I tell you this because I got more hung up watching that work than anything else.

As for a date- I am not sure. We talked about 10 minutes ago, and he thanked me for the results. And he said he would be offline for some time- I gathered that to mean more than just in terms of hours. But he did not elaborate. The issue of a release date has not come up. I know he has done extensive testing at speeds a lot faster than 1'+'1", and he wants something now a little more in line with the real world. I am not keeping any news hidden- you know as much about a release date right now as I do. Sorry I'm not much help here, right now.


Best regards,

gts