Page 2 of 28

Re: Chess.com 2018 computer chess championship

Posted: Sun Aug 19, 2018 1:00 pm
by zullil
Nay Lin Tun wrote:
Sun Aug 19, 2018 11:14 am
zullil wrote:
Sun Aug 19, 2018 9:50 am
Nay Lin Tun wrote:
Sun Aug 19, 2018 3:47 am
In September 3, 2018, chess.com will be hosting their computer chess championship for this year. There are a lot of background factors that favor Leela to be able to fight for No 1 vs the best AB engine , Stockfish. Of course, if you recently saw Leela performance in TCEC Division 3, this post will be a big joke( But in bonus games, she draw all 9 games vs Stockfish 8 and Ethereal).

Improvement factors/favoring factors for Leela,

1. 4x V 100 for Leela vs 48 cores ( GPU/CPU ratio will be 3.6 times in favor for Leela than TCEC, approximately + 90 elo)
2. Newer ID are already stronger than 520 (at least +50 elo)
3. Leela will have Tablebase support (possibly +20 elo)

As latest version of stockfish is only 100 elo stronger than stockfish 8, Leela will be in more favour if she can effectively use 4x V100 and Tablebases.

https://www.chess.com/computer-chess-championship
Is accurate information regarding hardware, time control and engine settings available online? Couldn't find anything at the URL above.
Here is another leaked link.
https://www.chess.com/news/view/announc ... ampionship
Thanks. I don't understand the choice to enable pondering. The server has 48 physical cores. Each engine will run 46 threads. So with pondering on, 92 threads will be running on only 48 physical cores. Seems like a bad idea to me.

Re: Chess.com 2018 computer chess championship

Posted: Sun Aug 19, 2018 2:18 pm
by kranium
zullil wrote:
Sun Aug 19, 2018 1:00 pm
Thanks. I don't understand the choice to enable pondering. The server has 48 physical cores. Each engine will run 46 threads. So with pondering on, 92 threads will be running on only 48 physical cores. Seems like a bad idea to me.
It's called 'hyperthreading' Louis, an Intel technology...each physical core = 2 logical cores
https://en.wikipedia.org/wiki/Hyper-threading

and it works great...anything you may have heard (for ex: no matter what, HT is 'bad' for chess engines) is a myth.
There's been plenty of tests and data presented here recently (by Kai? and others) showing it's beneficial.

As for our choice in the matter, it's unfortunate you don't agree. I'm sure there our thousands of weekend armchair tournament directors out there with a whole range of opinions on our choices, but I really don't intend to engage in a debate about them here. My intention was to simply provide information concerning the event.

Re: Chess.com 2018 computer chess championship

Posted: Sun Aug 19, 2018 2:21 pm
by zullil
kranium wrote:
Sun Aug 19, 2018 2:18 pm
zullil wrote:
Sun Aug 19, 2018 1:00 pm
Thanks. I don't understand the choice to enable pondering. The server has 48 physical cores. Each engine will run 46 threads. So with pondering on, 92 threads will be running on only 48 physical cores. Seems like a bad idea to me.
It's called 'hyperthreading' Louis, an Intel technology...each physical core = 2 logical cores
https://en.wikipedia.org/wiki/Hyper-threading
I'm quite familiar with hyperthreading. My concerns remain. :D

Re: Chess.com 2018 computer chess championship

Posted: Sun Aug 19, 2018 2:30 pm
by kranium
zullil wrote:
Sun Aug 19, 2018 2:21 pm
kranium wrote:
Sun Aug 19, 2018 2:18 pm
zullil wrote:
Sun Aug 19, 2018 1:00 pm
Thanks. I don't understand the choice to enable pondering. The server has 48 physical cores. Each engine will run 46 threads. So with pondering on, 92 threads will be running on only 48 physical cores. Seems like a bad idea to me.
It's called 'hyperthreading' Louis, an Intel technology...each physical core = 2 logical cores
https://en.wikipedia.org/wiki/Hyper-threading
I'm quite familiar with hyperthreading. My concerns remain. :D
Viewership is optional :D

Re: Chess.com 2018 computer chess championship

Posted: Sun Aug 19, 2018 2:45 pm
by zullil
kranium wrote:
Sun Aug 19, 2018 2:30 pm
zullil wrote:
Sun Aug 19, 2018 2:21 pm
kranium wrote:
Sun Aug 19, 2018 2:18 pm
zullil wrote:
Sun Aug 19, 2018 1:00 pm
Thanks. I don't understand the choice to enable pondering. The server has 48 physical cores. Each engine will run 46 threads. So with pondering on, 92 threads will be running on only 48 physical cores. Seems like a bad idea to me.
It's called 'hyperthreading' Louis, an Intel technology...each physical core = 2 logical cores
https://en.wikipedia.org/wiki/Hyper-threading
I'm quite familiar with hyperthreading. My concerns remain. :D
Viewership is optional :D
Of course. I understand that pondering is needed to "allow viewers to dive into each engine's thinking, as all engines will display live principal variations (PVs)—a real-time peek into their thinking process and the lines they are considering."

I have no interest in seeing PVs, since only the first few moves have any real value. The ever-changing displays of leaf positions that will never occur are simply distracting to me.

Good luck with the tournament.

Re: Chess.com 2018 computer chess championship

Posted: Sun Aug 19, 2018 4:11 pm
by jhellis3
I'm quite familiar with hyperthreading. My concerns remain.
Indeed.... Ponder is bad enough when threads < physical cores, but with using hyperthreads contention is guaranteed.

The real danger is not in CPU though, but concurrent TB access.... unless each program gets its own SSD with its own copy of whatever TBs it is using... which I highly doubt....

The bizarre thing is I know there are many people who do this, and they can not be persuaded that there is anything wrong with it. But whatever, not my electricity or money....

Re: Chess.com 2018 computer chess championship

Posted: Sun Aug 19, 2018 6:39 pm
by Laskos
jhellis3 wrote:
Sun Aug 19, 2018 4:11 pm
I'm quite familiar with hyperthreading. My concerns remain.
Indeed.... Ponder is bad enough when threads < physical cores, but with using hyperthreads contention is guaranteed.

The real danger is not in CPU though, but concurrent TB access.... unless each program gets its own SSD with its own copy of whatever TBs it is using... which I highly doubt....

The bizarre thing is I know there are many people who do this, and they can not be persuaded that there is anything wrong with it. But whatever, not my electricity or money....
They should use only 5-men Syzygy, otherwise it's a serious issue with ponder on.

Re: Chess.com 2018 computer chess championship

Posted: Sun Aug 19, 2018 6:47 pm
by zullil
jhellis3 wrote:
Sun Aug 19, 2018 4:11 pm
I'm quite familiar with hyperthreading. My concerns remain.
Indeed.... Ponder is bad enough when threads < physical cores, but with using hyperthreads contention is guaranteed.

The real danger is not in CPU though, but concurrent TB access.... unless each program gets its own SSD with its own copy of whatever TBs it is using... which I highly doubt....

The bizarre thing is I know there are many people who do this, and they can not be persuaded that there is anything wrong with it. But whatever, not my electricity or money....
SSD: 2x Crucial MX300 (1TB) in RAID1

https://www.chess.com/news/view/announc ... ampionship

Re: Chess.com 2018 computer chess championship

Posted: Sun Aug 19, 2018 6:52 pm
by kranium
Laskos wrote:
Sun Aug 19, 2018 6:39 pm
jhellis3 wrote:
Sun Aug 19, 2018 4:11 pm
I'm quite familiar with hyperthreading. My concerns remain.
Indeed.... Ponder is bad enough when threads < physical cores, but with using hyperthreads contention is guaranteed.

The real danger is not in CPU though, but concurrent TB access.... unless each program gets its own SSD with its own copy of whatever TBs it is using... which I highly doubt....

The bizarre thing is I know there are many people who do this, and they can not be persuaded that there is anything wrong with it. But whatever, not my electricity or money....
They should use only 5-men Syzygy, otherwise it's a serious issue with ponder on.
Hi Kai-
If you think TB access may be an issue, I'll seriously consider limiting to 5 man...
but I must say, we've been running this exact config (46 threads/ponder-on, and 6-man TBs) all summer with no issues.
Syzygy TBs can be accessed simultaneously by multiple engines.
Can you explain the issue further for me, and is there data to support it?

Re: Chess.com 2018 computer chess championship

Posted: Sun Aug 19, 2018 7:14 pm
by Laskos
kranium wrote:
Sun Aug 19, 2018 6:52 pm
Laskos wrote:
Sun Aug 19, 2018 6:39 pm
jhellis3 wrote:
Sun Aug 19, 2018 4:11 pm
I'm quite familiar with hyperthreading. My concerns remain.
Indeed.... Ponder is bad enough when threads < physical cores, but with using hyperthreads contention is guaranteed.

The real danger is not in CPU though, but concurrent TB access.... unless each program gets its own SSD with its own copy of whatever TBs it is using... which I highly doubt....

The bizarre thing is I know there are many people who do this, and they can not be persuaded that there is anything wrong with it. But whatever, not my electricity or money....
They should use only 5-men Syzygy, otherwise it's a serious issue with ponder on.
Hi Kai-
If you think TB access may be an issue, I'll seriously consider limiting to 5 man...
but I must say, we've been running this exact config (46 threads/ponder-on, and 6-man TBs) all summer with no issues.
Syzygy TBs can be accessed simultaneously by multiple engines.
Can you explain the issue further for me, and is there data to support it?
If there are 2 SSD, access is not a problem, so should be fine. The RAM usage might still be a problem, although I see a huge 256GB total available. With 6-men I do have problems in matches of engines which are not re-started even at short TC, but I have only 16GB available. Anyway, check the RAM usage with ponder on, but 256GB seems very safe.