SmarThink 1.40 updated

[Dayffd]

1. Minor fix (version info)
2. Added executables for x64/SSE3, x32

Available here: http://genes1s.net/smarthink.html

"The requested URL /smarthink.html was not found on this server."

Fixed

Thank you!

[Werewolf]

What did you change / improve from 1.2 ?

[Sergei S. Markoff]

What did you change / improve from 1.2 ?

Well, let's try to create a short list...

1. New lazy eval
2. Material imbalance tables update
3. Turn same-threat extension off
4. Several optimizations to speed up engine
5. Rehash steps 3 → 4
6. Store some move generator flags in TT to speed-up search
7. Tuning of king attack eval
8. Better move ordering: tuning history heuristic, bad-captures rehabilitation table
9. Detect some checkmates in quiescence (using eval data)
10. More accurate delta pruning
11. Refine null-move conditions/reductions
12. Refine IID conditions/params
13. Removed "quick-refutation" in history-based pruning
14. Code cleanup

[Ryan Benitez]

What did you change / improve from 1.2 ?

Well, let's try to create a short list...

1. New lazy eval
2. Material imbalance tables update
3. Turn same-threat extension off
4. Several optimizations to speed up engine
5. Rehash steps 3 → 4
6. Store some move generator flags in TT to speed-up search
7. Tuning of king attack eval
8. Better move ordering: tuning history heuristic, bad-captures rehabilitation table
9. Detect some checkmates in quiescence (using eval data)
10. More accurate delta pruning
11. Refine null-move conditions/reductions
12. Refine IID conditions/params
13. Removed "quick-refutation" in history-based pruning
14. Code cleanup

Do you have any new conclusions about same-threat (Botvinnik-Markoff) extensions? I have had mixed results trying it in the past, mostly breaking even but I have read that others like Tord got positive results from it at one time. Maybe it no longer fits in to the current trend of heavy reductions?

[Sergei S. Markoff]

Do you have any new conclusions about same-threat (Botvinnik-Markoff) extensions? I have had mixed results trying it in the past, mostly breaking even but I have read that others like Tord got positive results from it at one time. Maybe it no longer fits in to the current trend of heavy reductions?

I've just realised that turning STE off helps current SmarThink version while it worked for older versions. I have no explanation at the moment. May be it's because of the new pins/piece attacks evaluation, may be because newer SmarThink versions are turning null-move off for eval < beta nodes, may be it interacts with some other stuff.

[Werner]

Hi Sergei,
SmarThink normally uses 25% CPU on my Quad.
Now I watched Windows Task Manager and saw that every time when the engine starts calculating the Task Manager for 1 second shows 30-50% CPU and then drops back to 25%. Can you please explain it?

Maybe this is the reason of some crashes I have here when I parallel make 4 matches on the Quad ??

regards
Werner

PS: we make an update of our lists today with SmarThink 1.40 x64 avx Version.

[Sergei S. Markoff]

Hi Sergei,
SmarThink normally uses 25% CPU on my Quad.
Now I watched Windows Task Manager and saw that every time when the engine starts calculating the Task Manager for 1 second shows 30-50% CPU and then drops back to 25%. Can you please explain it?

Maybe this is the reason of some crashes I have here when I parallel make 4 matches on the Quad ??

regards
Werner

PS: we make an update of our lists today with SmarThink 1.40 x64 avx Version.

Absolutely no idea
25% means 100% of one core in the case of quad.
I have quad too and SmarThink never crashes on it. I'm using Fritz11 shell.
I never tried parallel matches and I think that it can in theory produce some concurrence problems if you're trying to use the same copy of the engine or the same copy of values.txt for each of engines copy...

[Werner]

Absolutely no idea
25% means 100% of one core in the case of quad.
I have quad too and SmarThink never crashes on it. I'm using Fritz11 shell.
I never tried parallel matches and I think that it can in theory produce some concurrence problems if you're trying to use the same copy of the engine or the same copy of values.txt for each of engines copy...

Hi Sergei,
thanks for the answer, but I think, you can watch your Task Manager too to see if you have the same behaviour..
With parallel games I do not use same engine twice. When I make 4 matches on the quad I make a game with Smarthink, another with Crafty, with Cheng and with Rodent e.g.
Now 100 % of CPU is used. But if now Smarthink uses more than 25% - this will cause Problems.
At the Moment I make only 3 matches on my quad - so I can see what happens when Smarthink starts calculating.
regards
Werner

[Graham Banks]

Thanks, Sergei.

Can you, please, link with static libraries ?
... so that we don't have to go downloading virus-infected dll files from internet.

Matthias.

I hope this dll is in redistributable package, so you can download it from Intel site.
As I've understood Intel compiler XE 2013 has some problems with static linking. If you will send me command line options I will compile it for you.

I'd like to start testing SmarThink 1.40 64-bit on my Intel i5, but I'm not sure where to get the extra files that I need.
Currently I get the missing file messages.

[carldaman]

Thanks, Sergei.

Can you, please, link with static libraries ?
... so that we don't have to go downloading virus-infected dll files from internet.

Matthias.

I hope this dll is in redistributable package, so you can download it from Intel site.
As I've understood Intel compiler XE 2013 has some problems with static linking. If you will send me command line options I will compile it for you.

I'd like to start testing SmarThink 1.40 64-bit on my Intel i5, but I'm not sure where to get the extra files that I need.
Currently I get the missing file messages.

I had the same problem, but it went away after I downloaded the dll package (w-ccompxe-redist-msi) and added the dlls to the ST1.40 folder. James I describes something similar above.

Edit: Here's the Intel download link for the w-ccompxe-redist-msi :

http://software.intel.com/en-us/article ... or-windows

Regards,
CL