Colossus Chess 2021 !!

[Gabor Szots]

Awesome!

There seems to be a few UCI / GUI issues using Colossus 2021 with the Shredder GUI. They PV in analysis mode is graded out, implying pondering. And the depth is shown as "0/20" i.e. fixed depth is always zero.

Looking forward to playing around with it.

Steve

It does not ponder under Arena (based on CPU usage) but it only displays the best move, no PV.

[MartinBryant]

Awesome!

There seems to be a few UCI / GUI issues using Colossus 2021 with the Shredder GUI. They PV in analysis mode is graded out, implying pondering. And the depth is shown as "0/20" i.e. fixed depth is always zero.

Looking forward to playing around with it.

Steve

Ah... sorry about that Steve. I've never used Shredder GUI.
I'll message you and we'll see if we can resolve it offline.

[mclane]

Awesome!

There seems to be a few UCI / GUI issues using Colossus 2021 with the Shredder GUI. They PV in analysis mode is graded out, implying pondering. And the depth is shown as "0/20" i.e. fixed depth is always zero.

Looking forward to playing around with it.

Steve

It does not ponder under Arena (based on CPU usage) but it only displays the best move, no PV.

You must set ponder on in the UCI parameters. By default it is IMO off.
Indeed the output is a little strange.
Under arena it shows search depth always with „1/„ before the depth.

But that is imo no big problem.

So far it runs ok, only i had the feeling king attacks of the opponent and opponents passed pawns are slightly underestimated.

[MartinBryant]

Awesome!

There seems to be a few UCI / GUI issues using Colossus 2021 with the Shredder GUI. They PV in analysis mode is graded out, implying pondering. And the depth is shown as "0/20" i.e. fixed depth is always zero.

Looking forward to playing around with it.

Steve

It does not ponder under Arena (based on CPU usage) but it only displays the best move, no PV.

You must set ponder on in the UCI parameters. By default it is IMO off.
Indeed the output is a little strange.
Under arena it shows search depth always with „1/„ before the depth.

But that is imo no big problem.

So far it runs ok, only i had the feeling king attacks of the opponent and opponents passed pawns are slightly underestimated.

Damn. Yes there does appear to be a couple of problems in Arena. Possibly the same issue Steve has in Shredder GUI. Although the analysis looks fine in Cutechess.
'Ponder' seems to be broken. It used to work in 2008b but I guess it's been broken somewhere in the last 13 years. Typically I use Cutechess-cli to run engine vs engine tests so haven't spotted these UI issues. Apologies! I can feel a 2021b version coming real soon!

[Wolfgang]

....

First of all: thanks for the new version after such a long time!!
CEGT 40/4 started: https://cegt.forumieren.com/t1581-testi ... 2021a#3059

We have 2460 for v2008b (we use another start rating than CCRL), so I expect something between 2550 and 2600 at our scale

1200 games played
Provisional rating: 2574 / +114!
Great work, Martin!

Concerning Ponder: some weeks ago I started a bigger tournament with engines below 2700 CEGT-ELO for our 5'+3"PB-list here:
https://cegt.forumieren.com/t1574-testi ... -2700#3046

Colossus 2008b is participating and makes no problems. Last place in the tournament has nothing to do with that...

Would be nice, if you could fix the ponder issue, I'd love to see 2021a (or however it is named then) in this list!

[MartinBryant]

....

First of all: thanks for the new version after such a long time!!
CEGT 40/4 started: https://cegt.forumieren.com/t1581-testi ... 2021a#3059

We have 2460 for v2008b (we use another start rating than CCRL), so I expect something between 2550 and 2600 at our scale

1200 games played
Provisional rating: 2574 / +114!
Great work, Martin!

Concerning Ponder: some weeks ago I started a bigger tournament with engines below 2700 CEGT-ELO for our 5'+3"PB-list here:
https://cegt.forumieren.com/t1574-testi ... -2700#3046

Colossus 2008b is participating and makes no problems. Last place in the tournament has nothing to do with that...

Would be nice, if you could fix the ponder issue, I'd love to see 2021a (or however it is named then) in this list!

That's good news. Thank you.
Yes, 2008b pondered ok.
I can see more than one place where I've broken it in 2021a so it's not a one line fix and it's also one of those things that needs thorough testing but it will return in 2021b.
By the way, I had no idea that you ratings list guys ever did 'ponder' testing

[Wolfgang]

....
By the way, I had no idea that you ratings list guys ever did 'ponder' testing

We started our 5'+3" list in 2013 , when Ingo Bauer temporarily stopped his great "IPON" list (finally abandoned in 2019).
https://web.archive.org/web/20190123144 ... inwoba.de/

In 2017 we started another Ponder-List with 3'+1".

5+3: http://www.cegt.net/rating5plus3pbon.htm
3+1: http://www.cegt.net/rating3plus1pbon.htm

[Colin-G]

I have recently installed linux on a couple of old computers to replace WinXP 32bit so I can run 64bit chess engines.
The standard version of Colossus2021a fails on both due to the old cpus, one intel, the other amd.
I tried the software popcount version which it said was 10% slower.
It works ok with Xboard using wine, but crashes whenever it sees a position in my 5 man Syzygy tablebases during its search.
It crashes instantly when loading a 5 man test position and choosing analysis mode.
If I remove the path to the Syzygy bases in my polyglot file it works fine.
It was the same on both computers.
I just completed a 10 game test match (40 moves in 4 mins) between Colossus2008b (with 5 man Nalimov tablebases) and Colossus2021a-software-popcount (without Syzygy)
This version of 2021a beat v2008b by 7:3 (no draws) on the above amd computer.

[MartinBryant]

I have recently installed linux on a couple of old computers to replace WinXP 32bit so I can run 64bit chess engines.
The standard version of Colossus2021a fails on both due to the old cpus, one intel, the other amd.
I tried the software popcount version which it said was 10% slower.
It works ok with Xboard using wine, but crashes whenever it sees a position in my 5 man Syzygy tablebases during its search.
It crashes instantly when loading a 5 man test position and choosing analysis mode.
If I remove the path to the Syzygy bases in my polyglot file it works fine.
It was the same on both computers.
I just completed a 10 game test match (40 moves in 4 mins) between Colossus2008b (with 5 man Nalimov tablebases) and Colossus2021a-software-popcount (without Syzygy)
This version of 2021a beat v2008b by 7:3 (no draws) on the above amd computer.

Hi Colin. Sorry to hear you're having a problem with it but I have no way to try your setup here.
Colossus uses the standard Ronald de Man probing code for the SYZYGY databases.
Do you have any other engines that use SYZYGY that you could try on your setup of Linux/Wine/Xboard?
If they don't work it would indicate the probing code.
If they do it would indicate my Colossus code.
Let me know.

[Colin-G]

I've not noticed any problems before when running various Windows engines that use Syzygy with wine in linux even on these old computers.
I just tried running running both standard 2021a and the sofware-popcnt versions on one of my newer linux computers. Both worked fine in Xboard and the windows version of Arena using wine. Syzygy use had no problems.
I had not seen any problems using either version with Windows 10 which is on my newer computers.
I will probably use v2021a for future engine matches instead of v2008b but not on those 2 old computers.
I have 532 games played by v2008b accumulated over the last 11 years, v2007c was used previously.
About half of my engine-engine matches are now played on linux computers using native linux versions where possible.