TCEC Tournament 4 - Long TC - Live - Now with Ivanhoe ++

[Martin Thoresen]

What happened in game 18 between Naum-Ivanhoe? (I think it is where the torunament stopped) I don't understand why the 1-0 result if Ivanhoe is evaluating it with a 0.00 and Naum with a 0.54. Rybka and Ivanhoe are so close that this game could decide the tournament.
Great work BTW.

Hello,

Ivanhoe crashed and thus lost on time.

; { Game Nr. 18 : Naum 4.2 64-bit 6CPU wins against Ivanhoe B52aF 64-bit 6CPU on time
; GameDuration = 05:44:42, Finalposition 5rk1/4r3/2R4Q/5q1P/4p2K/1P1p2P1/PR5P/8 b - - 3 51 }

[Vael Jean-Paul]

Then i would for sure reply the game!

JP.

[Martin Thoresen]

Then i would for sure reply the game!

JP.

No, as stated in the TCEC rules - if an engine lose on time by various reasons, the result will stand.
It just proves that Ivanhoe isn't without bugs.

I did replay a game once, before making this rule, and it created a small controversy.
Now that I have the rules in place beforehand I just want to obey by them.

[Robert Houdart]

Hi Martin,

On your web site is written that you've enabled Large Pages for engines that support this feature.
If both engines are capable of using Large Pages, it is quite possible that the OS provides sufficient Large Page memory to the first engine, but is unable to provide it to the second because of adressing space fragmentation.
How do you guarantee that the first engine that is loaded by the GUI doesn't have an unfair advantage?

Regards,
Robert

[Martin Thoresen]

Hi Martin,

On your web site is written that you've enabled Large Pages for engines that support this feature.
If both engines are capable of using Large Pages, it is quite possible that the OS provides sufficient Large Page memory to the first engine, but is unable to provide it to the second because of adressing space fragmentation.
How do you guarantee that the first engine that is loaded by the GUI doesn't have an unfair advantage?

Regards,
Robert

Hello Robert,

Large Pages aren't enabled yet (will be after TCEC Tournament 4 which is still running, and this info can also be found on the info page).

With 12 GB of RAM there isn't a problem running two engines with Large Pages enabled.

PS: How's Houdini doing? Still aiming for a December release?

[Robert Houdart]

With 12 GB of RAM there isn't a problem running two engines with Large Pages enabled.

In my experience (with 8 GB of RAM) Large Pages work quite well when the computer has just booted, but after a while the memory space becomes too fragmented for Windows to be able to allocate large chunks of contiguous memory. So it might be possible that only one of the engines will successfully obtain its Large Page memory.

PS: How's Houdini doing? Still aiming for a December release?

If all goes well, yes, - but that's a big if .

Cheers,
Robert

[Martin Thoresen]

Okay, I'll do some tests first to see if it works OK.

How large have you set the hash size and have you used 1 or 2 engines? I will be using 2 GB of hash per engine (there was a typo on my site which said 3 GB hash).
I have about 11 GB free RAM after windows is fully loaded, with only 24 processes running (yes I've disabled almost all non-vital services).

Glad to hear about Houdini, let me be the first to tell you that it will face R4 again in TCEC Match XX when it's released.

PS: Would be great if you could make an exclusive 6 thread build for the new version as I had big problems using the 8-thread build of 1.03a with 6 threads.

In my experience (with 8 GB of RAM) Large Pages work quite well when the computer has just booted, but after a while the memory space becomes too fragmented for Windows to be able to allocate large chunks of contiguous memory. So it might be possible that only one of the engines will successfully obtain its Large Page memory.

If all goes well, yes, - but that's a big if .

Cheers,
Robert

[Robert Houdart]

How large have you set the hash size and have you used 1 or 2 engines? I will be using 2 GB of hash per engine (there was a typo on my site which said 3 GB hash).
I have about 11 GB free RAM after windows is fully loaded, with only 24 processes running (yes I've disabled almost all non-vital services).

On my 8 GB server (running a lot of non-chess related applications) I only manage to get a 2 GB chunk *immediately* after booting - and that's just for a single engine. Your dedicated chess hardware should behave much better, your tests will tell...

Glad to hear about Houdini, let me be the first to tell you that it will face R4 again in TCEC Match XX when it's released.
PS: Would be great if you could make an exclusive 6 thread build for the new version as I had big problems using the 8-thread build of 1.03a with 6 threads.

Looking forward to the TCEC match . The plan is to release a single build of Houdini containing all the code for optimal operation from 1 to 8 cores, we'll make sure that it runs without problems on your hardware.

[Martin Thoresen]

Robert,

Seems you were right I cannot get both Rybka and Ivanhoe to use Large Pages at once, with a hash of 2048 MB. With a hash of 1024 MB each, it's no problem however.
So I have a hard time deciding what to do now, I really want to run with Large Pages but then only 1024 MB hash can be used for each engine.

What do you think would give the fastest depth overall? More hash or Large Pages?

If I am not going to run Large Pages I might as well set the hash to 4096 for both engines in this match.

[ernest]

What do you think would give the fastest depth overall? More hash or Large Pages?

Well, good question!
Not sure how much faster search 4GB hash is, compared to 1GB hash (I think nobody knows, and a meaningful test seems hard to achieve).
On the other hand, testing the speed increase with Large Pages is very easy.
and +5% is +5 Elo, +10% is +10 Elo , +15% is +14 Elo.