lkaufman wrote:I'm running a similar match to yours (details differ but fairly similar; slower machine, but one hour plus 30", 12 cores, smaller hash). My results are nearly the same; right now score is tied after 31 games.
But I'm getting better four core results than you did; so far Komodo is + 2 after 30 games compared to your -3 result for 50 games. But I think you used the initial release on that test while I used the newer bugfix one, so it's not exactly the same. Maybe the bugfix version is a bit stronger, and/or it's just sample error.
Thanks again for running these matches!
Best regards,
Larry
It in fact really is stronger, perhaps as much as 1 Elo see this tournament http://www.talkchess.com/forum/viewtopi ... 38&t=48468
this will account for 0,15 % better performance so your statement "maybe is a bit stronger" is true
But true also is that Komodo 5.1MP in this last match on 11 cores is suffering against Houdini
lkaufman wrote:I'm running a similar match to yours (details differ but fairly similar; slower machine, but one hour plus 30", 12 cores, smaller hash). My results are nearly the same; right now score is tied after 31 games.
But I'm getting better four core results than you did; so far Komodo is + 2 after 30 games compared to your -3 result for 50 games. But I think you used the initial release on that test while I used the newer bugfix one, so it's not exactly the same. Maybe the bugfix version is a bit stronger, and/or it's just sample error.
Thanks again for running these matches!
Best regards,
Larry
It in fact really is stronger, perhaps as much as 1 Elo see this tournament http://www.talkchess.com/forum/viewtopi ... 38&t=48468
this will account for 0,15 % better performance so your statement "maybe is a bit stronger" is true
But true also is that Komodo 5.1MP in this last match on 11 cores is suffering against Houdini
Hi
I guess the Houdini pro version has some advantages compared to the Standard version. The NUMA seems to do well on dualsocket platforms.
I started a new match, I hope its not boring that I choose Komodo and Houdini3 again. The last match was verry entertaining and had 25 wins out of 50 !!
So now I want to see how both do on more cores. Both engines get 11 cpu and 4GB hashtables this time. 1 cpu I need for live relay.
Conditions:
-Intel 12core machine X5680 @ 3.4GHz, no hyperthreading (2cpu on MoBo)
-4096 Mb hashtables, not EGTB at all
-GUI use small set of Nalimov EGTB (80Gb)
-Hiarcs book, 5 moves, tournament mode off(wide play)
-60min +10sec, ponder OFF
lkaufman wrote:I'm running a similar match to yours (details differ but fairly similar; slower machine, but one hour plus 30", 12 cores, smaller hash). My results are nearly the same; right now score is tied after 31 games.
But I'm getting better four core results than you did; so far Komodo is + 2 after 30 games compared to your -3 result for 50 games. But I think you used the initial release on that test while I used the newer bugfix one, so it's not exactly the same. Maybe the bugfix version is a bit stronger, and/or it's just sample error.
Thanks again for running these matches!
Best regards,
Larry
It in fact really is stronger, perhaps as much as 1 Elo see this tournament http://www.talkchess.com/forum/viewtopi ... 38&t=48468
this will account for 0,15 % better performance so your statement "maybe is a bit stronger" is true
But true also is that Komodo 5.1MP in this last match on 11 cores is suffering against Houdini
Hi
I guess the Houdini pro version has some advantages compared to the Standard version. The NUMA seems to do well on dualsocket platforms.
C.K.
Good point. I got an even result on 12 cores, and your result of just +2 for Houdini so far is very similar. So considering the NUMA support for Houdini, I guess we should be happy with such close results on more than six cores.
FWCC wrote:Nice comeback from Komodo.Was anything done to contribute to this improvement such as a reboot or something?
FWCC
Chess For Life
I don't thing anything was done to help Komodo if that is what you mean.
You should not look at short term trends in the win loss cycle, but the overall results. And the number of games played.
When two programs are playing that are close in strength. As I suspect with Komodo and Houdini 3, it is no different then flipping a coin. One side can get lucky for a time, but that luck will turn and even out with enough games.
IMO It would have been more surprising if this did not happen, and Houdini 3 showed itself to be even stronger on 11 CPU vs Komodo then on 4 CPU at the same time control.
"The worst thing that can happen to a forum is a running wild attacking moderator(HGM) who is not corrected by the community." - Ed Schröder
But my words like silent raindrops fell. And echoed in the wells of silence.
FWCC wrote:Nice comeback from Komodo.Was anything done to contribute to this improvement such as a reboot or something?
FWCC
Chess For Life
The matches are designed to run till the end without any touch.
Only in case of crashes or other strange happenings I restart / reboot or whatelse is needed to continue.
From time to time I look at the TasK Manger to see if hash and cpu usage is correct.
the match is over. Congratulations to Houdini 3 for the win(2 points ahead).
All in all it was a verry close match, with many decided games.
thx for your interests
best regards, Clemens Keck
PS.: Houdini 3 was running at split depht 12 in this match. I did several autotune tests, and it showed that SD 12 is best.
Hugo wrote:Hi All
I started a new match, I hope its not boring that I choose Komodo and Houdini3 again. The last match was verry entertaining and had 25 wins out of 50 !!
So now I want to see how both do on more cores. Both engines get 11 cpu and 4GB hashtables this time. 1 cpu I need for live relay.
Conditions:
-Intel 12core machine X5680 @ 3.4GHz, no hyperthreading (2cpu on MoBo)
-4096 Mb hashtables, not EGTB at all
-GUI use small set of Nalimov EGTB (80Gb)
-Hiarcs book, 5 moves, tournament mode off(wide play)
-60min +10sec, ponder OFF
