TalkChess.com

Cfish's nps was more than 3x that of Stockfish? (Looking at "Time elapsed", maybe not.)

I suppose the difference in local/remote ratio is due to Cfish making sure that the memory for each thread is allocated on the node on which the thread runs.

Nordlandia wrote: ↑

Sat Jun 22, 2019 12:21 pm

Hardware is assumed to be equal.

And I suppose you will be running the engines on separate machines (or the pondering engine will compete for resources with the non-pondering engine when it is the latter's turn).

Any insights? What happens when you store MOVE_NONE? Does that delete the old tt move (if there was one for the current position)? It is correct not to store the new "best" move if you haven't actually found a move that is better than start_alpha, but you should not delete the old tt move if there ...

Spliffjiffer wrote: ↑

Wed Jun 19, 2019 12:58 am

i see that the bug was found and already fixed (as far as i understood)...im glad that i could help a tiny bit in terms of verifying when to search...this should close the thread, thx for the replys

Thanks for reporting!

Any insights? What happens when you store MOVE_NONE? Does that delete the old tt move (if there was one for the current position)? It is correct not to store the new "best" move if you haven't actually found a move that is better than start_alpha, but you should not delete the old tt move if there ...

syzygy wrote: ↑

Tue Jun 18, 2019 7:45 pm

But none of the June 9 patches seem to relate to the relevant code.

I overlooked that this patch makes changes to "tbRank" in search.cpp and search.h. So that must be the problem. Should be happening in 6-piece positions too, I guess.

So with multiPV=5 Stockfish is showing 5 moves that lose, lose, draw, lose, lose in that order? That's pretty weird... It is correct that Kc5 draws and all other moves lose: https://syzygy-tables.info/?fen=2b5/8/3k3p/P7/8/3K1BP1/8/8_b_-_-_0_1 Of course it should not play a move that is known to lose...

Now my question is: what are strategies for finding what causes this slow down? The threads share no common variables apart from the transposition table. That tt has no locks, it uses the xor-trick. Most likely the threads share common cache lines, which has the same effect as sharing common variab...

G. Schlieb referred Shannon's groundbreaking paper, which is surely the base his paper. It has the same board represention (same piece codes), evaluation function and similar search ideas concerning selectivity and quiescence. And these are by far not the only similarities between the two papers. O...

Now my question is: what are strategies for finding what causes this slow down? The threads share no common variables apart from the transposition table. That tt has no locks, it uses the xor-trick. Most likely the threads share common cache lines, which has the same effect as sharing common variab...