On-line engine tourney March

[nionita]

I don't know but for 32 threads I would expect at least something like 35000 knps with Stockfish. Was this really 32 threads or was it hyperthreading or crippled AMD threads?

It's Opteron, 2x16 cores, so no hyperthreading. The only point is if xboard gave the engine all 32 threads - yesterday, by the first test games, I saw that xboard options had priority, the engine played with exactly the threads from xboard configuration, not the ones from polyglot configuration.

So when I corrected the xboard options, then it had 32 threads. But today (with the new command line options for xboard) I forgot to check the setting again. Although, with 1 thread Stockfich had only about 1 Mnps, so it can't be that.

I will do some test next days to see what's going on, with dirrerent number of threads and memory.

[Joost Buijs]

How strange can it be, Nightmare beat Stockfish but could not get more than a draw against RuyDos. It all depends upon the opening, RuyDos played with a good book and I had the impression that Stockfish played without book.

Yes, credit goes to José Manuel Morán, my book preparer and chess expert. Actually, the book is taken directly from Ruy-López, our program from the 90s.

Nightmare has some weaknesses in its book. The Marshall attack is not fully implemented. These openings are very difficult for an engine to comprehend. Anyway I am glad it did not lose because I saw it lose on some occasions to Thinker who also likes to play that variation with black.

[AlvaroBegue]

Are the games available for download somewhere?

[michiguel]

I don't know but for 32 threads I would expect at least something like 35000 knps with Stockfish. Was this really 32 threads or was it hyperthreading or crippled AMD threads?

In your experience, is the scaling in AMD worse? I am asking because I had different experience with quads and hexas.

Miguel

[Modern Times]

Cheng did rather well here !!

[Joost Buijs]

I don't know but for 32 threads I would expect at least something like 35000 knps with Stockfish. Was this really 32 threads or was it hyperthreading or crippled AMD threads?

In your experience, is the scaling in AMD worse? I am asking because I had different experience with quads and hexas.

Miguel

I have no experience with computers > 12 cores, and no experience with scaling on recent AMD boxes. On a 12 core Intel box Stockfish does something like 15000 knps, I would expect it to be at least twice this number for 32 cores.

On the other hand AMD cores are less efficient, about 70% of an Intel core, and the 32 core box probably runs at a much lower clock frequency. So it is very difficult to predict how fast it will be. However 9000 knps seems way to low.

Most multiprocessor boxes are NUMA instead of SMP, this can have a huge impact if your program is not specially prepared for it. I have no idea how Stockfish handles this.

[hgm]

Are the games available for download somewhere?

The games can be requested from the server, but not for the tournament as a whole, but only individually. The command is

mailstored PLAYER GAMENR

whre you can get the game numbers by

hist PLAYER

This will make the server e-mail the game to you.

Joshua Shriver used to run a bot that would collect all tourney games, but he was not there last night.