Page 14 of 23

Re: TCEC 10

Posted: Wed Nov 22, 2017 3:03 am
by shrapnel
Geonerd wrote: Setting contempt too high, relative to the other engines, can cost unnecessary losses, while setting it too low costs wins.
Well, as of now, it seems that Robert Houdart's gamble paid off !
Actually, using Contempt 2 was a Master Stroke by Houdart. If Houdini lost, he could always claim that using Contempt hurt Houdini ; if it won, nothing like it.
Personally, I've never been in favor of using Contempt 0, as it assumes that Chess is a perfectly balanced game.
But Chess, like Life, is seldom balanced.
Chess is both a Science and an Art, and it is the Art part that confounds both Human and Machine, as it cannot be quantified.
That is why, no World Champion, no matter how good his understanding of Chess, is ultimately beaten by someone else.
Quite often, playing a little aggressively and not necessarily the best move, puts the opponent (human or machine) off-balance and he/it also makes mistakes.
Of course, the key word is 'little' and going overboard with the Contempt is also a bad idea.

Re: TCEC 10

Posted: Wed Nov 22, 2017 1:26 pm
by Jouni
Is TCEC server 44 Cores -> 2 x Intel Xeon E5 2699 v4 @ 2.8 GHz numa? Maybe explanation to Houdini superior speed. And why isn't hyperthreading activated (88 threads would be nice)?

Re: TCEC 10

Posted: Wed Nov 22, 2017 2:22 pm
by Dann Corbit
Jouni wrote:Is TCEC server 44 Cores -> 2 x Intel Xeon E5 2699 v4 @ 2.8 GHz numa? Maybe explanation to Houdini superior speed. And why isn't hyperthreading activated (88 threads would be nice)?
Hyperthreading seems to make engines get more NPS on Epyc and Threadripper but not so much on Intel. When you double an already high core count you have way more than twice as much SMP loss also.

So two issues:
Does not really help NPS and much heavier SMP loss (which is already quite high due to the big core count)

Re: TCEC 10

Posted: Wed Nov 22, 2017 6:13 pm
by shrapnel
Hyperthreading seems to make engines get more NPS on Epyc and Threadripper but not so much on Intel
I have both i7 and AMD Ryzen and it is my considered opinion after a lot of Testing and Gaming, that HT On does NOT help either Intel or AMD.
The results may not be any worse, but they are certainly not any better.

Re: TCEC 10

Posted: Thu Nov 23, 2017 12:06 am
by Nay Lin Tun
Houdini leads with + 3 = 9 - 0 !!!

Re: TCEC 10

Posted: Thu Nov 23, 2017 2:41 am
by acase
shrapnel wrote:
Hyperthreading seems to make engines get more NPS on Epyc and Threadripper but not so much on Intel
I have both i7 and AMD Ryzen and it is my considered opinion after a lot of Testing and Gaming, that HT On does NOT help either Intel or AMD.
The results may not be any worse, but they are certainly not any better.


Thank you Anil, for confirming something I suspected a long time ago. I have always believed that "Hyperthreading" for chess is more about Hype rather than threading (those virtual threads don't really help chess programs much, I guess you could call it "virtual elo" :lol: )

Re: TCEC 10

Posted: Thu Nov 23, 2017 6:28 am
by AdminX
Nay Lin Tun wrote:Houdini leads with + 3 = 9 - 0 !!!
Correction make that 4 white wins. :shock:

Re: TCEC 10

Posted: Thu Nov 23, 2017 9:55 pm
by AdminX
Chessbase Review of cureent TCEC Season 10

https://en.chessbase.com/post/tcec-supe ... -vs-komodo

Re: TCEC 10

Posted: Thu Nov 23, 2017 11:24 pm
by AdminX

Re: TCEC 10

Posted: Fri Nov 24, 2017 8:02 am
by Dokterchen
Hi Ted,

thanks for sharing the links.

Torsten