CCRL Discussion Board

i just released new version (1.2) of Fizbo engine. It can be downloaded from https://sites.google.com/site/fizbochessengine/.

In this version:
1. improved search increases strength by around +70 ELO
2. now Fizbo supports parallel search. Self play using 4 cores shows strength increase of around +90 ELO.

Then, estimated ELO of Fizbo 1.2 should be around 2,840/2,930 (1 CPU/4 CPUs).

This is the first time i implemented parallel search logic. It is activated by command "cores N". I don't know if anything else needs to be changed in the engine to enable parallel search in Xboard protocol. If something else is needed, please let me know and i will implement it.

Thanks, Youri.

Excellent news! Thanks.

How many cores can Fizbo use?

it can use 8 cores. But i only tried it on 4 cores, since i only have 4 core computer. And i would not expect 8 cores to be much better than 4, since parallel search algorithm is simple and should not scale well.

Youri.

ymatioun wrote:it can use 8 cores. But i only tried it on 4 cores, since i only have 4 core computer. And i would not expect 8 cores to be much better than 4, since parallel search algorithm is simple and should not scale well.

Youri.

Thanks.

I have a gauntlet all ready to run, but the non-popcount engine doesn't work for me.

Sorry about that - i don't know why they don't run.

I recompiled and uploaded new binaries - could you please try them again? If i had some switches accidentally set to nonportable values, that should be fixed now. I think now i did exactly the same thing i did for previous version.

Also, parallel search uses condition variables, which were introduced in Windows Vista, so it would not run on Windows XP anymore.

If you encounter any more problems, could you please tell me what the error message is?

Thanks, Youri.

Hi Youri,

still no joy on my Q8200.
There is no error message.
I doubleclick on the exe and a blank console window opens.
I create the engine under ChessGUI, but it either doesn't load (as task manager doesn't register it) or just uses up its time without displaying any output at all.

The popcount version is fine.

Sorry to be awkward,
Graham..

For what it's worth, the non-popcount exe runs fine on a popcount machine.

I don't know what is going on.

The only difference between "popcount" and "non popcount" versions is that instead of "__popcnt64()" manually coded version of that function is used in "_no_pc" executable.

I just tried running both files on my laptop and on my work computer, and they all seem to work fine. (run them, type "xboard", "new", "go", and see the search output).

Maybe this is due to parallel search logic? As i said before, the only interface command that triggers parallel search is xboard command "cores 4". If some interface requires something else, then it would not work.

What interface are you using? Let me try it and see if i observe any problems.

Also, if you say that "non pc" executable runs, then let's just go ahead with it.

Thanks, Youri.

Hi Youri,

I use the ChessGUI interface for my testing.

I usually run gauntlets on my Q8200 and Q6600 computers - both non-popcount.
I reserve the i5 (popcount) for running the Amateur Divisions.

I will try your suggestion (run them, type "xboard", "new", "go", and see the search output) and see what I get.

Graham.

let me look into this and see what i can find.

Thanks, Youri.

OK, i found the problem - 2 occurrences of "popcount" sneaked into the "no_pc" version.
I removed them, and verified by reviewing disassembly that there are no more of them in the "no_pc"executable.
Sorry about that - i did not carry the code to automatically replace all the popcounts from version 1.1.
But now it should run - please give it a try.

Thanks, Youri.

Works well now. Thanks Youri.

Gauntlet will start Monday or Tuesday with live broadcast in TLCV port 16066.

Great, thanks!