TCEC - Division 3 - Season 1 - Live! Discuss.

[BB+]

On multi CPU configurations though this is not an issue.

We have a 4 cpu 16-core machine at work. It has 128GB, and each cpu "controls" 32GB as it were (8 DIMMs of 4GB each), with requests to such a physical memory location needing to go through to controlling cpu. It is not the newest machine, and the HyperTransport between cpus is sufficiently slow that I have clocked as much as a 2x slowdown when the OS manages to put a process on one cpu while its memory accesses are largely controlled by another [I was running a 10-core 64GB process for awhile, and could not figure out why some of cores seemed to run almost twice as fast as others]. Admittedly, I don't think this a problem for most systems, but it's another pitfall that can occur.

[Martin Thoresen]

Thanks Richard and BB for the explanation.

As you may or may not know now the ponder off will stay.

Division III starts tomorrow!

[Martin Thoresen]

And Division III is now officially live!

[kingliveson]

I worry at times when you play Ponder-On games on the same machine...

Why is that?

The total CPU usage is still 100%, if that is 6 cores from 1 engine or 3 cores from 2 engines.

Anyway, there are some issues with ChessGUI with Ponder ON making it crash during saving of livepgn, so I will not use ponder on.

I didn't want to give a Hennessey and Patterson lecture on cache contention, so thought I would try to demonstrate the problem, but could not reproduce it. What was said above essentially are possible problems.

Anyways, so I was able to produce the problem in Arena. Load the engines, and hit analyze. When set to 1 CPU each on a quad-core, you wont see the problem, but when set to 2 cpus, then you see where potential problem can occur. See the pictures, engine on the bottom is always getting better kns though it's the same engine.

[Martin Thoresen]

kingliveson,

I am not at all a programmer so I didn't know this before you guys said it. No worries.

But what strikes me with your pictures, shouldn't you use the exact same build/binary when comparing?

[Matthias Gemuh]

kingliveson,

I am not at all a programmer so I didn't know this before you guys said it. No worries.

But what strikes me with your pictures, shouldn't you use the exact same build/binary when comparing?

I think it is same build/binary, but one must be renamed.

[kingliveson]

kingliveson,

I am not at all a programmer so I didn't know this before you guys said it. No worries.

But what strikes me with your pictures, shouldn't you use the exact same build/binary when comparing?

I think it is same build/binary, but one must be renamed.

Jim Ablett said recently that he would start using Microsoft compiler for 32-bit builds vs Intel, and PeterPan posted something similar the other day, so I was doing speed test of the two hence the "i" and "m" naming. But yes the files are builds from the same compiler -- didnt want to install another engine, so instead I just copied and overwrote the file with the same name.

[Martin Thoresen]

Standings after round 2 of 14:

Code: Select all

Nr Engine                               Pts      SB        A H J T U S G D
   1 Arasan 12.2                          6.0      2.0000  - - - - - 1 1 -
   2 Hannibal 1.0a                        4.0      4.0000  - - - 1 = - - -
   3 Jonny 4.0                            4.0      0.5000  - - - - - = - 1
   4 Tornado 4.4                          3.0      0.0000  - 0 - - - - - 1
   5 Umko 1.1                             2.0      2.5000  - = - - - - = -
   6 Scorpio 2.6                          1.0      2.0000  0 - = - - - - -
   7 Gaviota 0.8                          1.0      1.0000  0 - - - = - - -
   8 Daydreamer 1.75                      0.0      0.0000  - - 0 0 - - - -

[Dave Mitchell]

Nice to see Arasan doing well.

I prefer matches with ponder off, it's on the same system, because I'm always thinking that one engine could "cheat", and jump itself up to higher priority, or use more cpu time that it's opponent, by "working around" the ponder off settings.

I recall the old days, when engines used to do such dirty tricks as continually query the interface, to help prevent the opponent's engine, from getting it's full share of cpu cycles.

Only a very sophisticated (and cynical), user would ever notice this of course, making said engine appear "very strong", when being tested on the same system as any other program.

I understand that you have set ponder off. I hope the ChessGUI interface, will not allow such trickery.

[Martin Thoresen]

Dave,

Thanks for your input!

All engines are thoroughly tested in blitz and rapid before the main event starts, this is to ensure that nothing fishy is going on.
In this case, all engines behaved exemplary.

The current game Arasan - Jonny is quite interesting, seems Arasan has a shot at pulling ahead of the rest of the field.