Stockfish 8

2295 views
Skip to first unread message

mcos...@gmail.com

unread,
Nov 1, 2016, 5:31:01 AM11/1/16
to FishCooking, Joona Kiiski, Daylen Yang, Kiran Panditrao
Me and Joona decided to release SF 8 with the same sources that will partecipate in TCEC superfinal.

We think it is a sensible choice that makes SF 8 a well defined point in time release, not just a random commit.

In the following hours Daylen will update our official site.

Kiran will kindly compile the target binaries.

chess...@gmail.com

unread,
Nov 1, 2016, 6:19:21 AM11/1/16
to FishCooking, joona....@gmail.com, d...@daylenyang.com, pand...@gmail.com
Congratulations Sir.. and many thanks to the whole team of developers and testers! Go Stockfish !

graci...@gmail.com

unread,
Nov 1, 2016, 6:36:22 AM11/1/16
to FishCooking, joona....@gmail.com, d...@daylenyang.com, pand...@gmail.com

No regression test or smp test before official release? I don't know I feel this was rushed. I would at least wait for tcec result and the numa patch and keep up with old january official releases.

I expect houdini 5 to win now, and keep the conspiracy alive ;-)

pasquale....@gmail.com

unread,
Nov 1, 2016, 6:44:23 AM11/1/16
to FishCooking, joona....@gmail.com, d...@daylenyang.com, pand...@gmail.com

Kiran Panditrao

unread,
Nov 1, 2016, 9:20:39 AM11/1/16
to FishCooking, joona....@gmail.com, d...@daylenyang.com
Dear Sir,

Thanks. I express deep regrets for not been able to provide 'Linux' builds. My Linux machine is off.

However, here are 'Windows' builds :


All the windows builds are 'static'.The 32 bit one has been done on 64 bit machine, so can someone with 32 bit windows ensure compatibility please ? 

Sincere regards,
KP
Stockfish 8 Windows pl rename ext from abc to exe.zip

Ipman Chess

unread,
Nov 1, 2016, 9:25:08 AM11/1/16
to FishCooking, joona....@gmail.com, d...@daylenyang.com
No Kiran..your 32bit compile is a 64bit compile..

Op dinsdag 1 november 2016 14:20:39 UTC+1 schreef Kiran Panditrao:

Ipman Chess

unread,
Nov 1, 2016, 9:51:11 AM11/1/16
to FishCooking, joona....@gmail.com, d...@daylenyang.com
Stockfish 8 32 gen
Stockfish 8 32 old
Stockfish 8 32 sse
Stockfish 8 64 BMI2
Stockfish 8 64 POP
Stockfish 8 64 SSE

https://mega.nz/#!nQAHwIpR!YjhlLR602IpezEvZ__vGLUP8KK4a9dugkNdth7ae_c0

Enjoy your testings

Ipman.

Kiran Panditrao

unread,
Nov 1, 2016, 10:03:34 AM11/1/16
to FishCooking, joona....@gmail.com, d...@daylenyang.com
Thanks, i suspected it could be 64 bit. 

However, your builds are 'non-static', for official release Static ones are required. ("We are not going to start shipping dlls. Even though this might be a tiny bit faster, we would be flooded with queries and bug reports from inexperienced users. So please use static linking..." -Zamar)

Regards,
KP

pasquale....@gmail.com

unread,
Nov 1, 2016, 10:18:23 AM11/1/16
to FishCooking, joona....@gmail.com, d...@daylenyang.com
Kiran, follow the windows 32 bit instructions. You need only to download mingw-w64 for 32 bit

https://github.com/glinscott/fishtest/wiki/Building-stockfish-on-Windows#mingw-w64--msys
> Ipman.

apassionforcr...@gmail.com

unread,
Nov 1, 2016, 10:18:40 AM11/1/16
to FishCooking
Marco,

Do you think that it may be a good idea to give Anton or Martin new compiles from Kiran that are named "Stockfish 8"? That way during the TCEC match people can see the name Stockfish 8. I understand that this would be for only aesthetic purposes. But still - it would go down in TCEC archives as Stockfish 8. Also it has been confirmed by Martin that hyperthreading will be disabled; thus there should be no need at all for a special NUMA compile since the processor groups will still just be one.

Cheers,

-Apassion-

Kiran Panditrao

unread,
Nov 1, 2016, 10:28:47 AM11/1/16
to FishCooking, joona....@gmail.com, d...@daylenyang.com, pasquale....@gmail.com
Thanks, I am aware, but my net connection (2g mobile phone) did not permit me to do so in time. so i tried to play around a bit. ... well it couldn't have succeeded ... I apologize everyone for providing a 64 bit one in the name of 32 bit. ... sorry again ... 

Kiran Panditrao

unread,
Nov 1, 2016, 10:31:55 AM11/1/16
to FishCooking, apassionforcr...@gmail.com
The SF 8 is 'static' ... TCEC is LTO ...  it's faster than the one provided here ... But if maintainers decide, i can submit a LTO SF 8 ...

Ipman Chess

unread,
Nov 1, 2016, 10:33:16 AM11/1/16
to FishCooking, joona....@gmail.com, d...@daylenyang.com
Yes they are Static Kiran..no dlls needed ;)

Op dinsdag 1 november 2016 15:03:34 UTC+1 schreef Kiran Panditrao:

Marco Costalba

unread,
Nov 1, 2016, 10:33:58 AM11/1/16
to Kiran Panditrao, FishCooking, joona....@gmail.com, d...@daylenyang.com, pasquale....@gmail.com
Not a problem Kiran you did a great job. 

In case you think you can compile the 32 bit in the next days we can add them later. Please let us know. 

pasquale....@gmail.com

unread,
Nov 1, 2016, 10:37:36 AM11/1/16
to FishCooking, joona....@gmail.com, d...@daylenyang.com, pand...@gmail.com, pasquale....@gmail.com
Compile tests

Windows 64 bit
(Windows 10 64 bit - make profile-build ARCH=x86-64-modern COMP=mingw)

mingw-w64 6.2.0: PASS (ppigazzini)
mingw-w64 6.2.0 (Msys2): PASS (ppigazzini)
mingw-w64 5.4.0: PASS (ppigazzini)
mingw-w64 4.9.3: PASS (ppigazzini)
mingw-w64 4.8.5: PASS (ppigazzini)

Windows 64 bit - bash on ubuntu on windows
(make profile-build ARCH=x86-64-modern COMP=gcc)

gcc 4.8.4: PASS (ppigazzini)


Windows 32 bit
(Windows XP SP3 32 bit - make profile-build ARCH=x86-32 COMP=mingw)

mingw-w64 6.2.0: PASS (ppigazzini)


Windows cross compilation from linux
(make profile-build ARCH=x86-64-modern COMP=mingw)

Ubuntu 14.04 mingw-w64 4.8.2: PASS (ppigazzini)
Ubuntu 15.10 mingw-w64 4.9.2: PASS (ppigazzini)

(make build ARCH=x86-64-modern COMP=mingw)
Ubuntu 16.04 mingw-w64 5.3.1: PASS (ppigazzini)


Linux 64 bit
(make profile-build ARCH=x86-64-modern COMP=gcc)

Ubuntu 14.04 gcc 4.8.4: PASS (ppigazzini)
Ubuntu 15.10 gcc 5.2.1: PASS (ppigazzini)
Ubuntu 16.04 gcc 5.4.0: PASS (ppigazzini)
Ubuntu 16.10 gcc 6.2.0: PASS (ppigazzini)

joost.van...@gmail.com

unread,
Nov 1, 2016, 10:46:41 AM11/1/16
to FishCooking, joona....@gmail.com, d...@daylenyang.com, pand...@gmail.com, pasquale....@gmail.com

Ronald de Man

unread,
Nov 1, 2016, 11:16:35 AM11/1/16
to FishCooking
On Tuesday, November 1, 2016 at 3:31:55 PM UTC+1, Kiran Panditrao wrote:
The SF 8 is 'static' ... TCEC is LTO ...  it's faster than the one provided here ... But if maintainers decide, i can submit a LTO SF 8 ...


Does MingW not allow you to link statically with LTO?
I see no reason why those options should bite each other (which does not mean it actually works).

pasquale....@gmail.com

unread,
Nov 1, 2016, 11:39:26 AM11/1/16
to FishCooking
A old bug in Mingw-w64 for Windows (that affects all distributions: msys, msys2, tdm-gcc etc.) does not allow a static LTO. Or this was the situation when I tried some months ago.

Cross-compiling from Linux allow a LTO static build, but without the speedup that dynamic LTO provides on Windows.

(LTO is a proved boost only for CPU >= Haswell; with my Ivy Bridge I don't have any speedup)

Ronald de Man

unread,
Nov 1, 2016, 12:15:46 PM11/1/16
to FishCooking
On Tuesday, November 1, 2016 at 4:39:26 PM UTC+1, pasquale....@gmail.com wrote:
A old bug in Mingw-w64 for Windows (that affects all distributions: msys, msys2, tdm-gcc etc.) does not allow a static LTO. Or this was the situation when I tried some months ago.

Cross-compiling from Linux allow a LTO static build, but without the speedup that dynamic LTO provides on Windows.

(LTO is a proved boost only for CPU >= Haswell; with my Ivy Bridge I don't have any speedup)

I'm now verifying it myself. Static LTO indeed crashes for me too.

Apparently COMP=gcc (or leaving COMP unspecified) will give LTO, COMP=mingw will give static.

My own pgo LTO build seems slightly faster than the pgo static build on my Sandybridge laptop, but the difference is very small. (With Cfish LTO has always slowed down so far for me, but perhaps it is different on Haswell and later.)

Could be interesting to try to compile a pgo Windows binary on Linux. I don't know if it is possible, but I don't see why not (the pgo stats would have to be collected by running stockfish through wine, but why not). I'll briefly give it a shot.

pasquale....@gmail.com

unread,
Nov 1, 2016, 12:26:47 PM11/1/16
to FishCooking
Yes, LTO is a little speedup also for my Ivy Bridge (0.3%), on Haswell is a 2% speedup.

Instructions required (LTO, cross compile etc) on the wiki :)

https://github.com/glinscott/fishtest/wiki/Building-stockfish-on-Windows

Mindbreaker

unread,
Nov 1, 2016, 12:30:43 PM11/1/16
to FishCooking, joona....@gmail.com, d...@daylenyang.com, pand...@gmail.com
So, skipping the tablebase patches?  Seems like it would have been a good opportunity to have people check that it all works correctly being we can't test tables in the Framework.

Well, maybe it was all sufficiently tested already.

Ronald de Man

unread,
Nov 1, 2016, 12:49:30 PM11/1/16
to FishCooking
On Tuesday, November 1, 2016 at 5:15:46 PM UTC+1, Ronald de Man wrote:

Could be interesting to try to compile a pgo Windows binary on Linux. I don't know if it is possible, but I don't see why not (the pgo stats would have to be collected by running stockfish through wine, but why not). I'll briefly give it a shot.

Making a windows pgo build on Linux turns out to be as easy as making a windows build without pgo. It just works. Speed is practically the same, as expected.

Unfortunately making a static build on Linux has its own problems.

pasquale....@gmail.com

unread,
Nov 1, 2016, 1:00:24 PM11/1/16
to FishCooking
Cross-compilation already on the wiki :)

https://github.com/glinscott/fishtest/wiki/Building-stockfish-on-Windows#cross-compilation-from-ubuntu

Unfortunately Ubuntu 16.04 and Ubuntu 16.10 are only able to compile without PGO (bug in latest mingw-w64 package for Debian/Ubuntu)

-pp

ps: if you use Ubuntu try lxc/lxd

Daylen Yang

unread,
Nov 1, 2016, 1:34:24 PM11/1/16
to FishCooking, pand...@gmail.com, joona....@gmail.com, d...@daylenyang.com, pasquale....@gmail.com
I will update the Stockfish website with Kiran's Windows x64, x64 popcnt, and x64 bmi2 builds, and I'll upload my compiles for Mac soon. We're still missing the Windows 32-bit compile and Linux builds, which I assume Kiran can send later.

Daylen Yang

unread,
Nov 1, 2016, 1:56:24 PM11/1/16
to FishCooking, pand...@gmail.com, joona....@gmail.com, d...@daylenyang.com, pasquale....@gmail.com
The download page (https://stockfishchess.org/download/) has been updated with Kiran's Windows binaries. (Refresh if the page is cached.) I'll wait for Kiran to send the Linux binaries and I'll upload the Mac binaries soon. (Android binaries haven't been updated in a while...)

pasquale....@gmail.com

unread,
Nov 1, 2016, 2:15:04 PM11/1/16
to FishCooking, pand...@gmail.com, joona....@gmail.com, d...@daylenyang.com, pasquale....@gmail.com
Abrok's Linux binaries are fine.

We need only the windows 32 bit build from Kiran.

-pp

Marco Costalba

unread,
Nov 1, 2016, 2:56:40 PM11/1/16
to Pasquale Pigazzini, FishCooking, Kiran Panditrao, Joona Kiiski, Daylen Yang
Aren't the Abrok build much slower?

Marco Costalba

unread,
Nov 1, 2016, 2:56:54 PM11/1/16
to Daylen Yang, FishCooking, Kiran Panditrao, Joona Kiiski, Pasquale Pigazzini
Great! thanks Daylen.

Kiran Panditrao

unread,
Nov 1, 2016, 3:42:20 PM11/1/16
to FishCooking, joona....@gmail.com, d...@daylenyang.com, pasquale....@gmail.com, Marco Costalba
Here is the 32 bit Windows binary. 


Also attached here for convenience.

I once again express deep regrets that I will not be able to provide Linux builds. My Linux machine is off and won't be up so soon.

Regards,
KP
stockfish_8_x32 rename ext from abc to exe.zip

Marco Costalba

unread,
Nov 1, 2016, 3:52:23 PM11/1/16
to Kiran Panditrao, FishCooking, joona....@gmail.com, d...@daylenyang.com, pasquale....@gmail.com
Not a problem and thanks for the binaries. 

Daylen,  we can grab the abrok ones for Linux binaries (anyhow interested people will probably compile by themselves). 

2424dyn...@gmail.com

unread,
Nov 1, 2016, 4:07:01 PM11/1/16
to FishCooking, joona....@gmail.com, d...@daylenyang.com, pand...@gmail.com
Is there going to be a test to compare SF8 to SF7 on 1,3, and 15 threads just like we did when you released SF7?

Marco Costalba

unread,
Nov 1, 2016, 4:14:09 PM11/1/16
to 2424dyn...@gmail.com, FishCooking, joona....@gmail.com, d...@daylenyang.com, pand...@gmail.com
We did a regression test just few days ago. 

Regarding the official elo difference with sf7 I leave to public lists to find.  Our self match with our book at one minute per game is anyhow not very representative of the final public outcome. 

Daylen Yang

unread,
Nov 1, 2016, 4:51:29 PM11/1/16
to Marco Costalba, 2424dyn...@gmail.com, FishCooking, joona....@gmail.com, pand...@gmail.com
I've updated the Windows zip on the site with the 32-bit binary from Kiran, and zipped up the abrok.eu Linux binaries and put those up too.

Eric Mullins

unread,
Nov 1, 2016, 5:00:05 PM11/1/16
to FishCooking
In this thread (https://groups.google.com/forum/#!topic/fishcooking/exhr8eW2pps) my suggestion to have SF8 be the same as the TCEC superfinal wasn't particularly well recieved.  I'm glad to see that it did happen this way after all, though.  Thanks.

stefano.c...@gmail.com

unread,
Nov 1, 2016, 5:08:28 PM11/1/16
to FishCooking, joona....@gmail.com, d...@daylenyang.com, pand...@gmail.com
On Tuesday, November 1, 2016 at 10:31:01 AM UTC+1, mcos...@gmail.com wrote:
> Me and Joona decided to release SF 8 with the same sources that will partecipate in TCEC superfinal.
>
> We think it is a sensible choice that makes SF 8 a well defined point in time release, not just a random commit.
>
> In the following hours Daylen will update our official site.
>
> Kiran will kindly compile the target binaries.

Maybe a blog entry?

Marco Costalba

unread,
Nov 1, 2016, 5:46:36 PM11/1/16
to Daylen Yang, 2424dyn...@gmail.com, FishCooking, joona....@gmail.com, pand...@gmail.com
Thanks Daylen!

So I have just made the official announcement on takchess:

http://www.talkchess.com/forum/viewtopic.php?p=691852#691852


pasquale....@gmail.com

unread,
Nov 1, 2016, 6:08:15 PM11/1/16
to FishCooking, pasquale....@gmail.com, pand...@gmail.com, joona....@gmail.com, d...@daylenyang.com
@Marco: Abrok's Windows cross compiles are slower than native ones: it's not possible to perform a PGO build because a bug in Ubuntu 6.04 mingw-w64 package.

Abrok's Linux compiles are fine.
Obviously cherry picking a time based PGO build can allow a 0.1% speedup :)

-pp

fkr...@gmail.com

unread,
Nov 1, 2016, 6:25:00 PM11/1/16
to FishCooking, joona....@gmail.com, d...@daylenyang.com, pand...@gmail.com
So how much elo did SF gain from 7 -> 8?

compute...@gmail.com

unread,
Nov 1, 2016, 6:50:13 PM11/1/16
to FishCooking, joona....@gmail.com, d...@daylenyang.com, pand...@gmail.com
On Tuesday, November 1, 2016 at 9:31:01 AM UTC, mcos...@gmail.com wrote:
> Me and Joona decided to release SF 8 with the same sources that will partecipate in TCEC superfinal.
>
> We think it is a sensible choice that makes SF 8 a well defined point in time release, not just a random commit.
>
> In the following hours Daylen will update our official site.
>
> Kiran will kindly compile the target binaries.


The x64 popcount exe from https://stockfishchess.org/ does not run on my AMD X6 1090T, it crashes. This CPU does have popcnt support.

But the Stockfish modern exe "Stockfish 301016 64 POPCNT" from ultimaiq.net does work on the X6 1090t and that I believe is exactly Stockfish 8. So effectively I have a popcnt version of SF8 for the X6 now even though not strictly "official".

apassionforcr...@gmail.com

unread,
Nov 1, 2016, 6:51:35 PM11/1/16
to FishCooking
Probably about 80-82 Elo. Something like that.

Vah

unread,
Nov 1, 2016, 8:59:20 PM11/1/16
to FishCooking
Stockfish Popcnt from off site is also crashing on x4 965. I guess Kiran made some tampering with makefile

Kiran Panditrao

unread,
Nov 2, 2016, 12:03:23 AM11/2/16
to FishCooking, joona....@gmail.com, d...@daylenyang.com, compute...@gmail.com, Marco Costalba
Oops. I had tested them on Intel machines. Worked fine. Also tested again on 'intel', works fine ... unfortunately i did not have an AMD machine to test ...

It's my mistake again. I truly feel sorry and shameful again, I let the SF team down again ... my bad ...

However, here are the builds that work on ALL systems ...


Also attached here for convenience.

Sincere regards and apologies again ...
KP
Stockfish 8 win change ext from abc to exe.zip

Daylen Yang

unread,
Nov 2, 2016, 12:59:40 AM11/2/16
to Kiran Panditrao, FishCooking, joona....@gmail.com, compute...@gmail.com, Marco Costalba
No worries, I just replaced the 64-bit and popcnt Windows binaries on the site with the ones you just sent.

Kiran Panditrao

unread,
Nov 2, 2016, 2:46:55 AM11/2/16
to FishCooking
Thanks !!

Aprijal Pasaribu

unread,
Nov 2, 2016, 5:18:05 AM11/2/16
to FishCooking, joona....@gmail.com, d...@daylenyang.com, pand...@gmail.com


Pada Selasa, 01 November 2016 16.31.01 UTC+7, mcos...@gmail.com menulis:
Me and Joona decided to release SF 8 with the same sources that will partecipate in TCEC superfinal.

We think it is a sensible choice that makes SF 8 a well defined point in time release, not just a random commit.

In the following hours Daylen will update our official site.

Kiran will kindly compile the target binaries.


HERE for android builds. 

Adam Herwis

unread,
Nov 2, 2016, 11:21:24 PM11/2/16
to FishCooking, joona....@gmail.com, d...@daylenyang.com, pand...@gmail.com


On Tuesday, November 1, 2016 at 5:31:01 PM UTC+8, mcos...@gmail.com wrote:
Me and Joona decided to release SF 8 with the same sources that will partecipate in TCEC superfinal.

We think it is a sensible choice that makes SF 8 a well defined point in time release, not just a random commit.

In the following hours Daylen will update our official site.

Kiran will kindly compile the target binaries.




I think the release is kinda premature, perhaps reschedule to January or at least after the TCEC championship. 

Nabil Danial

unread,
Nov 3, 2016, 12:05:21 AM11/3/16
to FishCooking, joona....@gmail.com, d...@daylenyang.com, pand...@gmail.com
Please update the blog on Tumblr. Some stats would be nice, i.e. lines added/deleted vs SF 7.

cwsto...@gmail.com

unread,
Nov 3, 2016, 7:24:03 AM11/3/16
to FishCooking, fkr...@gmail.com
Am Dienstag, 1. November 2016 23:25:00 UTC+1 schrieb fkr...@gmail.com:
> So how much elo did SF gain from 7 -> 8?

Andreas Strangmüller ran a test of 8000 games with TC = 20" + 0.2" with a 86 ELO gain as result:

1 Stockfish 8 64 POPCNT : 3043 5 5 8000 62.1 % 2957 62.7 %
2 Stockfish 7 64 POPCNT : 2957 5 5 8000 37.9 % 3043 62.7 %

More details to be found here:

http://talkchess.com/forum/viewtopic.php?topic_view=threads&p=692024&t=61930

Daylen Yang

unread,
Nov 3, 2016, 5:30:18 PM11/3/16
to Nabil Danial, FishCooking, joona....@gmail.com, pand...@gmail.com
Nabil, I have just updated the blog with that information: http://blog.stockfishchess.org/post/152699558142/stockfish-8 
Message has been deleted

pado...@gmail.com

unread,
Nov 3, 2016, 9:36:34 PM11/3/16
to FishCooking, nabilda...@gmail.com, joona....@gmail.com, pand...@gmail.com
1500 GAME GAUNTLET
i3 380M | 2 @ 2.53 | Win7 x64 | 6 GB Ram | Gui~ CuteChess GUI 0.9.4
Draw Adjudication~120 / 40 / 2 | Resign~3 / 4.8 | Cores~1 HT~On
Book~TopGM_2move.pgn | Hash~8MB | Concurrency~3
Ponder~No | TB's~No | Elo~Ordo 1.2.6 | Time~20" + .2" 

Stockfish 8 +70 Elo over Komodo 10.2 

Code:
   # PLAYER         :  RATING  ERROR  PLAYED    W    D    L   (%)  D(%)
   1 Stockfish 8    :    3235      7    1500  619  559  322  59.9  37.3
   2 Komodo 10.2    :    3165      7    1500  322  559  619  40.1  37.3

hadda...@gmail.com

unread,
Feb 16, 2018, 7:56:35 AM2/16/18
to FishCooking
Hi
Please send me an email (personal email) to haddad.zied (at) gmail.com

I have a business proposal.
Thanks

pado...@gmail.com

unread,
Feb 16, 2018, 10:09:25 AM2/16/18
to FishCooking
Who are you talking to?
Reply all
Reply to author
Forward
0 new messages