Komodo 10.4 vs Houdini 5

[mjlef]

Yes, you are right, there were over 100 games. But still you need a lot more than that to reach any meaningful comparison.

As for MP scaling, how do you know what we have done since 10.3?

I do not mean to be argumentative. I just thing more data would make claims more meaningful.

[beram]

Yes, you are right, there were over 100 games. But still you need a lot more than that to reach any meaningful comparison.

As for MP scaling, how do you know what we have done since 10.3?

I do not mean to be argumentative. I just thing more data would make claims more meaningful.

In the releasenotes of Komodo 10.3 you wrote about ''improvement for MP performance''. In the last for Komodo 10.4 there is nothing mentioned about that

As for the claim this was first made by Larry and I gave examples of data who proove otherwise

Also there isn't yet any testmatch result between Komodo 10.4 and Houdini 5 which subscribe the +1 ELO 1thread at 2m1s (+13 on 4threads) above Houdini 5 given by Larry
For now it looks that you will need just one other release to equalize with Houdini 5

FGRL 60 0,6: Houdini 5 x64 : 250 ( 43, 143, 64), 45.8%
Ipon: Komodo 10.4 - Houdini 5.01 (3285) 105.0 - 115.0 47.73%
CEGT: Houdini 5.0 x64 1CPU 3322 91.5-108.5 45,75%

[JJJ]

On fast.gm Komodo 10.4 is still below Houdini 5 as well on 1 minute control and 1CPU.

But this is the worst time control for Komodo. I d prefer to wait result at 10 min time control. We might see Komodo equal to Houdini 5 here.

[lkaufman]

Yes, you are right, there were over 100 games. But still you need a lot more than that to reach any meaningful comparison.

As for MP scaling, how do you know what we have done since 10.3?

I do not mean to be argumentative. I just thing more data would make claims more meaningful.

In the releasenotes of Komodo 10.3 you wrote about ''improvement for MP performance''. In the last for Komodo 10.4 there is nothing mentioned about that

As for the claim this was first made by Larry and I gave examples of data who proove otherwise

Also there isn't yet any testmatch result between Komodo 10.4 and Houdini 5 which subscribe the +1 ELO 1thread at 2m1s (+13 on 4threads) above Houdini 5 given by Larry
For now it looks that you will need just one other release to equalize with Houdini 5

FGRL 60 0,6: Houdini 5 x64 : 250 ( 43, 143, 64), 45.8%
Ipon: Komodo 10.4 - Houdini 5.01 (3285) 105.0 - 115.0 47.73%
CEGT: Houdini 5.0 x64 1CPU 3322 91.5-108.5 45,75%

Komodo 10.4 came out 4 elo above Houdini 5.01 on the IPON list. The total rating is of course vastly more significant statistically than the individual match result. Also no results yet for four threads, which is where I got the statistically significant plus score. But the exact results do seem to be fairly hardware-dependent; just between my Intel xeon machines relative speeds and results vary noticeably.

[beram]

First match results 100games on my AMD x6 1090T
TC 3m2s,both 4cores, ponder off, privat book 25 lines
A small margin win for Komodo 10.4 +16 -14 =70
A next 100 games is running

On my i5 laptop 2cores same engines, same TC, I have after 75 games
+12 -8 =55 for Houdini 5

Conclusions sofar Komodo has improved its results, it looks on par with Houdini 5. Off course more results needed to confirm

Code: Select all

Komodo10.4 - Houdini5 4x Blitz 3m+2s
                        
1   Komodo 10.4 64-bit  +16/=70/-14 51.00%   51.0/100
2   Houdini 5 x64-popc  +14/=70/-16 49.00%   49.0/100

Match with 5x times longer time control on my i5 laptop 2cores
at TC 15m10s, after first cycle of 50 games same pattern 51% very close
+7 -6 =37 for Houdini 5

Code: Select all

K10.4 - H5, Rapid 15m+10s                              
1   Houdini 5 x64-pext    +7  +7/=37/-6 51.00%   25.5/50
2   Komodo 10.4 64-bit    -7  +6/=37/-7 49.00%   24.5/50

[Amateur]

give me komodo engine and I will test on my 2x Intel Xeon E5-2673 v4...

[Milos]

give me komodo engine and I will test on my 2x Intel Xeon E5-2673 v4...

You could afford a dual configuration of CPUs that even as qualification sample cost like 1300$ per unit and you can't afford Komodo???? Yea sure

[Amateur]

You could afford a dual configuration of CPUs that even as qualification sample cost like 1300$ per unit and you can't afford Komodo???? Yea sure

it is already in testing, but only 14 cpus for engine, time 5' and with Perfect 2017 by Sedat as opening book ...

[Scacchista1977]

I'm conducting a similar test with i7 5960x, at moment 80 games +14 elo for Houdini. In total will be 500 games.

[Scacchista1977]