SF6 has been released

[styx]

Thanks to Daylen, official binaries are now available for download in:
http://stockfishchess.org/download/

We expect these to be somewhat faster than the ones abrok.eu is providing. However there should be no need to restart test runs or anything like that if you already have used abrok.eu compiles.

hi. the official android version(sf 6 JA) is based on arm5 compile from JA so its slower than arm7 compile,any date for arm7 compile?

You can try this version: https://dl.dropboxusercontent.com/u/896 ... ish-arm.gz. I don't know it if will become the official version. On my HTC One it is about 20% faster than the JA version.

Thanks Peter just what I was looking for, faster than the ja version.

The official android download has been updated now, so no need to use my link anymore.

do you have any idea why your android-compiled SF6 is about 10% slower than the dev-version of SF that comes with the latest droidfish app? at least on my phone

[cdani]

Hi. This seems a bug on stockfish 6:

Code: Select all

Stockfish 6 64 POPCNT by Tord Romstad, Marco Costalba and Joona Kiiski
position fen r1q1nr1k/pp1b2b1/n2p2pp/2pP1p2/2B4B/3Q1N1P/PPP1NPP1/1R3RK1 b - - 0
12
go depth 15
info depth 1 seldepth 1 multipv 1 score cp -33 nodes 31 nps 31000 tbhits 0 time
1 pv g6g5
info depth 2 seldepth 2 multipv 1 score cp 609 nodes 116 nps 116000 tbhits 0 tim
e 1 pv a6b4 c2c3 b4d3
info depth 3 seldepth 3 multipv 1 score cp 609 nodes 169 nps 169000 tbhits 0 tim
e 1 pv a6b4 c2c3 b4d3
info depth 4 seldepth 4 multipv 1 score cp 601 nodes 258 nps 258000 tbhits 0 tim
e 1 pv a6b4 e2f4 b4d3 c4d3
info depth 5 seldepth 5 multipv 1 score cp 523 nodes 545 nps 545000 tbhits 0 tim
e 1 pv a6b4 e2f4 b4d3 c4d3 g6g5 f4g6 h8g8 g6f8
info depth 6 seldepth 8 multipv 1 score cp 324 nodes 878 nps 878000 tbhits 0 tim
e 1 pv a6b4 e2f4 b4d3 f4g6 h8h7 g6f8 g7f8 c4d3
info depth 7 seldepth 9 multipv 1 score cp 338 nodes 1153 nps 104818 tbhits 0 ti
me 11 pv a6b4 e2f4 b4d3 f4g6 h8h7 g6f8 g7f8 c4d3 c5c4
info depth 8 seldepth 10 multipv 1 score cp 101 nodes 7725 nps 367857 tbhits 0 t
ime 21 pv g6g5 c4a6 b7a6 f3g5 h6g5 h4g5 e8f6 e2f4
info depth 9 seldepth 12 multipv 1 score cp 76 nodes 12335 nps 587380 tbhits 0 t
ime 21 pv a6b4 d3d2 g6g5 h4g5 h6g5 d2g5 b7b5 c2c3 e8f6
info depth 10 seldepth 15 multipv 1 score cp 56 nodes 25000 nps 609756 tbhits 0
time 41 pv a6b4 d3d2 g6g5 a2a3 b7b5 a3b4 b5c4 b4c5 c8c5 h4g5 h6g5 d2g5 c5d5
info depth 11 seldepth 16 multipv 1 score cp 47 nodes 67287 nps 1103065 tbhits 0
 time 61 pv g6g5 h4g5 a6b4 d3d2 h6g5 f3g5 b7b5 a2a3 b5c4 a3b4 c5b4 e2f4 f8f6 d2b
4
info depth 12 seldepth 18 multipv 1 score cp 64 nodes 103100 nps 1132967 tbhits
0 time 91 pv g6g5 h4g5 a6b4 d3d2 h6g5 d2g5 h8g8 e2f4 b7b5 c2c3 b5c4 c3b4 f8f7 b4
c5
info depth 13 seldepth 19 multipv 1 score cp 73 nodes 148466 nps 1226991 tbhits
0 time 121 pv g6g5 h4g5 a6b4 d3d2 h6g5 f3g5 b7b5 a2a3 b5c4 a3b4 c5b4 e2f4 f8f6 d
2b4 a8b8
info depth 14 seldepth 22 multipv 1 score cp 36 nodes 336177 nps 1394925 tbhits
0 time 241 pv g6g5 f3g5 h6g5 h4g5 e8f6 e2f4 h8g8 f4e6 f6e4 e6f8 g8f8 c4a6 e4g5 a
6c4 g5e4 c2c3 f8g8
info depth 15 seldepth 22 multipv 1 score cp 5 nodes 831105 nps 1455525 tbhits 0
 time 571 pv g6g5 h4g5 a6b4 d3d2 h6g5 f3g5 e8f6 a2a3 b7b5 e2f4 h8g8 c4b3 b4a6 f4
g6 c8e8 g6f8 e8f8
bestmove g6g5 ponder h4g5

So it drops the queen in the firsts depths.

[gladius]

Hi. This seems a bug on stockfish 6:

<snip>

So it drops the queen in the firsts depths.

It's not a bug unfortunately. Just before SF 6 a patch was introduced to do futility pruning in the PV. It only happens at low depths, but, yeah, it can lead to silly looking things like this.

On the plus side, this should be some incentive to improve the futility pruning rules.

[drowsy]

In the linux package there are 3 binaries, x64, x64_bmi2, and x64_modern. Could you please update the readme file to explain what they are?

Thanks!

PS I found the info via google, but it should be in the readme file with the distribution, imho.

[Sylwy]

In the linux package there are 3 binaries, x64, x64_bmi2, and x64_modern. Could you please update the readme file to explain what they are?

Thanks!

PS I found the info via google, but it should be in the readme file with the distribution, imho.

for old CPUs
for Haswell architecture CPUs
for modern CPUs (supporting sse 4.2 & popcnt instruction sets)

SilvianR

[Dr.Wael Deeb]

In the linux package there are 3 binaries, x64, x64_bmi2, and x64_modern. Could you please update the readme file to explain what they are?

Thanks!

PS I found the info via google, but it should be in the readme file with the distribution, imho.

When I find several executables in a package I try them all and use the one that does not crash and is the fastest. Am I primitive?

My mouth is sealed

[petero2]

hi. the official android version(sf 6 JA) is based on arm5 compile from JA so its slower than arm7 compile,any date for arm7 compile?

You can try this version: https://dl.dropboxusercontent.com/u/896 ... ish-arm.gz. I don't know it if will become the official version. On my HTC One it is about 20% faster than the JA version.

Thanks Peter just what I was looking for, faster than the ja version.

The official android download has been updated now, so no need to use my link anymore.

do you have any idea why your android-compiled SF6 is about 10% slower than the dev-version of SF that comes with the latest droidfish app? at least on my phone

When compiling using the android ndk-build script instead of using the stockfish Makefile, a lot of compiler flags are different, which in this case makes the binary about 35% smaller and 7% faster. I have not been able to modify the Makefile to produce the same binary, but here is the binary extracted from the DroidFish build:

https://dl.dropboxusercontent.com/u/89684995/sf6-arm.gz

[Leto]

The official binary seems to be between 10 to 14 percent faster than the abrok compile.

[matthewlai]

Congratulations! I think it is an important achievement. I believe in the idea of cooperation for the sake of advancing science without concern for monetary gain. It is, IMHO a high expression of the best of what we aspire to.

Thank you so much for your hard work! I love chess and computer chess holds an unending fascination for me. So you have made this old guy happier today.

I wish continued success for the whole Stockfish team. Also, kudos to all the testers who freely contributed their time and resources to the project.

Very nice!

I do not think that the target of the stockfish team is to advance science(otherwise instead of only trying to improve stockfish they can do tests to find out what are the main factors that help stockfish to be strong).

It can be done simply by removing every piece of code that does not give a significant elo gain and see what we get.

It is possible simply to test every code that we want to remove
with SPRT(-6,0) at 1 minute per game and remove it if it pass the test.

I guess that the final result is going to be weaker than stockfish6 but still better than stockfish5 and we may know better what are the main factors
that help stockfish to be a strong engine.

Stockfish is definitely contributing to science by being open source.

For example, I am doing some research using experimental approaches to play chess, and need a stable, well optimized, and clean chess engine as a starting point. Stockfish fits the bill perfectly.

[lech]

I was looking through the search of Stockfish and I see that Stockfish fights with "R" value in "null move search" like .... Tinapa ?
It is a good idea!
Who was the author?