Progress of Houdini 1.5

[Tomcass]

TESTING PROGRESS OF HOUDINI 1.5.

Computer: i7 975 4 real cores + 4 Hyper Thread. (HT : Off)
Fritz Benchmark: Speed 26,73 KNS 12.828
Gui: Fritz 12
Time Control: 2' + 1
Book: H.S. Masterbook 3.1
Ponder: Off.
No Robbo, Triple or Nalimov.
No crashes.

Last test : Houdini 1.5 against a selection of strong Ivanhoes. (400 games)

Houdini 1.5 x64 - Ivanhoe B50kBx64p 55.0 - 45.0 55.00%
Houdini 1.5 x64 - IvanHoe 9.49b x64CNT_KLO 54.0 - 46.0 54.00%
Houdini 1.5 x64 - Ivanhoe B48vAx64a 53.0 - 47.0 53.00%
Houdini 1.5 x64 - IvanhoeB50tAX64p 50.5 - 49.5 50.50%

(What a great result for IvanHoeB50tAx64p!!)

Average Houdini 1.5: 53.12%

Houdini 1.03a x64 - Ivanhoe B50kBx64p 46.5 - 53.5 46.50%
Houdini 1.03a x64 - IvanHoe 9.49b x64PopCNT_KLO 45.0 - 55.0 45.00%
Houdini 1.03a x64 - Ivanhoe B48vAx64a 46.5 - 53.5 46.50%
Houdini 1.03a x64 - IvanhoeB50tAX64p 47.5 - 52.5 47.50%

Average Houdini 1.03a: 46.37%

Deep Rybka 4 - Ivanhoe B50kBx64p 43.0 - 57.0 43.00%
Deep Rybka 4 - IvanHoe 9.49b x64PopCNT_KLO 45.0 - 55.0 45.00%
Deep Rybka 4 - Ivanhoe B48vAx64a 45.5 - 54.5 45.50%
Deep Rybka 4 - IvanhoeB50tAX64p 43.5 - 56.5 43.50%

Average Deep Rybka 4: 44.25%

Regards,

Tom.

[ernest]

As I already said, Tom, I wish I had found such results with the recent Ivans against DR4, but unfortunately I did not.
For me, the best Ivans are equal, not better than DR4.

Something really new has to be found with the Ivans, to make them better (like H15), not YAC (Yet Another Compile)...

Maybe somebody else can confirm (or not) your tests...

[Tomcass]

Ernest,

There is a very simple way for you to check these results. Test the mentioned IvanHoes (one thousand games would be enough) against Deep Rybka 4 yourself. As far as I know, more than a dozen of serious testers have done it and their results are very similar to those I have reported. Thousands of games have been played proving this. But obviously you are free to believe whatever you want.

At this time control, the difference between the best IvanHoes and DR4 is substantial. Not so clear at longer time controls.

Regards from Barcelona.

Tom.

[gerold]

As I already said, Tom, I wish I had found such results with the recent Ivans against DR4, but unfortunately I did not.
For me, the best Ivans are equal, not better than DR4.

Something really new has to be found with the Ivans, to make them better (like H15), not YAC (Yet Another Compile)...

Maybe somebody else can confirm (or not) your tests...

Still looking for that IH that can match Houdini. Like you my
test at 5/3 TC shows the strongest verisons of IH to be very
close to R4.

Best,
Gerold.

[Tomcass]

Computer: i7 975 4 real cores + 4 Hyper Thread. (HT : Off)
Fritz Benchmark: Speed 26,73 KNS 12.828
Gui: Fritz 12
Time Control: 2' + 1
Book: H.S. Masterbook 3.1
Ponder: Off.
No Robbo, Triple or Nalimov.
No crashes.

Latest test: Houdini 1.5 (14_2) against a selection of strong Ivanhoes. (400 games).

Houdini 1.5 x64(14_2)- Ivanhoe B50kBx64p 59.0 - 41.0 59.00%
Houdini 1.5 x64(14_2)- Ivanhoe B48vAx64a 55.5 - 44.5 55.50%
Houdini 1.5 x64(14_2)- IvanhoeB50tAX64p 53.0 - 47.0 53.00%
Houdini 1.5 x64(14_2)- IvanHoe 9.49b x64CNT_KLO 57.0 - 43.0 57.00%

Average Houdini 1.5 (14_2): 56.12%


Houdini 1.5 x64 - Ivanhoe B50kBx64p 55.0 - 45.0 55.00%
Houdini 1.5 x64 - IvanHoe 9.49b x64CNT_KLO 54.0 - 46.0 54.00%
Houdini 1.5 x64 - Ivanhoe B48vAx64a 53.0 - 47.0 53.00%
Houdini 1.5 x64 - IvanhoeB50tAX64p 50.5 - 49.5 50.50%
(What a great result for IvanHoeB50tAx64p!!)

Average Houdini 1.5: 53.12%

Houdini 1.03a x64 - Ivanhoe B50kBx64p 46.0 - 54.0 46.00%
Houdini 1.03a x64 - IvanHoe 9.49b x64PopCNT_KLO 45.0 - 55.0 45.00%
Houdini 1.03a x64 - Ivanhoe B48vAx64a 46.5 - 53.5 46.50%
Houdini 1.03a x64 - IvanhoeB50tAX64p 47.5 - 52.5 47.50%

Average Houdini 1.03a: 46.37%

Deep Rybka 4 - Ivanhoe B50kBx64p 43.0 - 57.0 43.00%
Deep Rybka 4 - IvanHoe 9.49b x64PopCNT_KLO 45.0 - 55.0 45.00%
Deep Rybka 4 - Ivanhoe B48vAx64a 45.5 - 54.5 45.50%
Deep Rybka 4 - IvanhoeB50tAX64p 43.5 - 56.5 43.50%

Average Deep Rybka 4: 44.25%

I think that my test with Houdini 1.5 SD14 Contempt 2 confirms the good results got by other serious testers. In my computer it seems to be somewhat stronger than the standard Houdini 1.5, at this time control and against IvanHoes. A bit more than 20 ELO points.

Regards,

Tom.