TalkChess.com

Look wrote:[
[...]

Why Stockfish 7 binaries and source code is not available from the official site http://stockfishchess.org/ ?

it is: http://stockfishchess.org/download/

zamar wrote:http://stockfishchess.org/download/ has been updated.
It contains the currently fastest known binaries for Stockfish 7.

Hi Joona,

I just downloaded the Stockfish 7 Windows binaries
from http://abrok.eu/stockfish and http://stockfishchess.org/download/

They are exactly the same !!!

Where are "the fastest binaries" ?

Really Dude? You received a decent gift and this is all you can say? Jim Ablett and the other compiler experts owe you nothing.

Compile it yourself if you think the abrok guy did a bad job.

Dude yourself !
I am just surprised...
Usually the official Stockfish is faster than the corresponding dev Stockfish.

Don't get a stroke for that !...

ernest wrote: Usually the official Stockfish is faster than the corresponding dev Stockfish.

I thought it was the other way around; mainly that the official Stockfish is more stable and the development versions are the ones where "what if" lines of thought are exercised.

Of course by faster you mean nps, than I have no idea.

I'm pretty sure SF6 had faster compliles than SF Dev.

Werewolf wrote:I'm pretty sure SF6 had faster compliles than SF Dev.

I can confirm that !

In SF forum: no one didn't beat abrok compiles more than popcnt + 0.5 % and x64 + 0.7 % so no need other compiles.
But official bmi compile is 4,6 % faster! BTW why are testers running popcnt version with corei7 CPU - no bmi instructions?

ernest wrote:

zamar wrote:http://stockfishchess.org/download/ has been updated.
It contains the currently fastest known binaries for Stockfish 7.

Hi Joona,

I just downloaded the Stockfish 7 Windows binaries
from http://abrok.eu/stockfish and http://stockfishchess.org/download/

They are exactly the same !!!

Where are "the fastest binaries" ?

Quoting Joona on Fishtest, there was an open challenge on building the fastest binaries, quickly after freezing the code so that the test groups could start immediately and would not have to wait. I believe I read something elsewhere on the newsgroup about the owner of Abrok having a new server so he did not have to crosscompile (from Linux presumably?).

https://groups.google.com/forum/?fromgr ... _mxqBDdwRg

Based on the results, so far, I'd say that:

- We'll go with abrok's builds
- Except for Windows x64 Hashwell where we'll use Kiran's static build.

Jouni wrote:In SF forum: no one didn't beat abrok compiles more than popcnt + 0.5 % and x64 + 0.7 % so no need other compiles.
But official bmi compile is 4,6 % faster! BTW why are testers running popcnt version with corei7 CPU - no bmi instructions?

I think the microcode was disabled for BMI in some of the earlier Haswells (?)