IvanHoe52hUS fron Vlad0

Vlad0 writes


Vladimir Molotkov
to me

show details 4:45 PM (35 minutes ago)


Hello girls
Yesterday I found the simple optimization of IVH missed somehow by decembrists.
And this night ran a standard battle of the new compile,
optimized (and with "no bug" fixed), with Houdini. The result

1 Ivanhoe 52hUS-x64 +9 +41/=123/-36 51.25% 102.5/200
2 Houdini 1.01 x64 2_CPU -9 +36/=123/-41 48.75% 97.5/200

was pleasant -- the win of IVH though a small, but with older compiles
it was not observed for a long time already. But it was unclear, whether this
result was due to the new optimization of IVH, or just a statistical fluctuation?

To check this I've ran a quicky 50 games bullet 10". The result here

1 Houdini 1.01 x64 2_CPU +28 +14/=26/-10 54.00% 27.0/50
2 Ivanhoe 52hUS-x64 -28 +10/=26/-14 46.00% 23.0/50

as well was a surprise. Though IVH was behind Houdini yet,
but in the similar match 2 days before, with non-optimized IVH, the result was much worse:

1 Houdini 1.01 x64 2_CPU +123 +25/=17/-8 67.00% 33.5/50
2 Ivanhoe 52gUS-x64 -123 +8/=17/-25 33.00% 16.5/50.

This shows that the new IVH is really quicker than the old one.
Which is confirmed by visual observation of games as well.
One can see also, that in bullet regime the time management of IVH
is worse than that of Houdini, though speeds of both engines are visually
the same, IVH much more often looses due to timeouts.

The new compile is attached, and now I'll download it
to Immortal's as well.

http://www.multiupload.com/TS063D2B4L


~~~~~~~~~~~~~~



confused by silly bullet 10 second games test we asked ....

is this engine stronger than 52gU ?


and support large pages ?

Vlad answered...


Definitely! It is just quicker "g" version with killed "no move" bug!

Of course! Nothing is removed from g!


Vlad0

From Vlad , also on his bog

IVH 52hUS

Posted Today at 20:19 by Vlad0
Updated Today at 20:22 by Vlad0
New features:

1) "No move" bug fixed, at last!
Though not in the best way yet, but it works!
Not a single "no bug" in 200 blitz 1+0 and
in 150 bullets 10" up to now. And the engine
is not slowed down by this fix.

2) Moreover, I found yesterday a way to
speed up the engine, overlooked by decembrists With this optimization even
in bullet match against Houdini the new engine
is not smashed so catastrophically as show
2 quicky bullet 10" examples:
New compile:

1 Houdini 1.01 x64 2_CPU +28 +14/=26/-10 54.00% 27.0/50
2 Ivanhoe 52hUS-x64 -28 +10/=26/-14 46.00% 23.0/50

Old compile:

1 Houdini 1.01 x64 2_CPU +123 +25/=17/-8 67.00% 33.5/50
2 Ivanhoe 52gUS-x64 -123 +8/=17/-25 33.00% 16.5/50.

And the first normal blitz 1+0 with new compile

1 Ivanhoe 52hUS-x64 +9 +41/=123/-36 51.25% 102.5/200
2 Houdini 1.01 x64 2_CPU -9 +36/=123/-41 48.75% 97.5/200

Of course one is to add about 1000 games more to see that this win is not just a fluctuation.

3) Now the engine autodetects the moment where each new game begins. Now it is not used for anything useful, but can be used
to clean deteriorated hash after, say, each 10 games or to keep in memory the previous game
and take info from any odd game and use it in the next even. If every opening opens twice
it may reduce thinking over the next move in case that the opponent has won the preceeding
game.

i have just started a new tournament in 10 minutes by games between this version and houdini 64 bits 2cpu.
result will come soon

where can we found the source corrected to compile it on my linux platform ?

Thanks
TheropodeS

theropodes wrote:

i have just started a new tournament in 10 minutes by games between this version and houdini 64 bits 2cpu.
result will come soon

where can we found the source corrected to compile it on my linux platform ?

Thanks
TheropodeS

go to ippolite wikie and send Vlad0 a pm , he may help you , all we can give you are the decemberists source codes

ok thank i will do it.

At this time here, the result

RITCHIE, Blitz 10m 0

With the bug
Houdini 1.03a x64 2_CPU - IvanHoe T0.4 beta2 6.0 - 4.0 +3/=6/-1 60.00%

Without the bug
Houdini 1.03a x64 2_CPU - Ivanhoe 52hUS-x64 3.5 - 2.5 +1/=5/-0 58.33%

The final result with a tournament 10 minutes between Houdini and Ivanhoe on a XP64 IntelCore2Duo 2.33

Blitz 10m
Houdini 1.03a x64 2_CPU - Ivanhoe 52hUS-x64 6.0 - 4.0 +3/=6/-1 60.00%

keep in touch to the right way, step by step Ivanhoe will win.

Regards
TheropodeS