Independent Computer Chess Discussion Forum
https://open-chess.org/
Thank you for that great idea! It had never crossed my mind, actually.Robert Houdart wrote:That was quick!
It works very well, thank you!
Robert
Of course, I always use the best build available.Vael Jean-Paul wrote:You gonne use Houdini 1.03a popcnt x64 version!!
Stockfish does use POPCNT, in fact the Stockfish executable contains both the standard and POPCNT versions of most routines. At start-up it detects whether the CPU supports POPCNT and subsequently uses the appropriate routines.Martin Thoresen wrote:If SF had a popcnt version I'd use it as well.
Thanks Robert, but your input on this matter makes me confused. I was talking to Tord at chessbomb.com chat during game 27, and I quote the whole conversation:Robert Houdart wrote:Stockfish does use POPCNT, in fact the Stockfish executable contains both the standard and POPCNT versions of most routines. At start-up it detects whether the CPU supports POPCNT and subsequently uses the appropriate routines.
For Houdini the POPCNT version is only marginally (say 2%) faster than the normal version, so it's not really a big deal.
Robert
greenchile505: Very interesting, thanks.
greenchile505: Tord, when you are testing a new rev of an engine, what engine do you play it against? The previous? or an engine from another author? or a human?
Xaurus: Tord I was asking Marco for a POPCNT build of SF, is that something you guys are considering? He said that it might not be faster than Jim's builds. In this match, Rybka is running with the POPCNT executable (SSE 4.2).
ChessBomb: Speaking of Jim's builds, Tord, do you know which compiler he uses under Linux? Our quick-and-dirty no-SSE42 builds are actually faster than his PGO build.
jhoro: hi Xaurus, does Rybka use tablebases?
dennis1382: <Tablebases: 3-4-5 Men Nalimov>
TordRomstad: @greenchile505: Normally, we just play self-play matches against the previous version.
Xaurus: jhoro, yes all 3, 4 and 5 men tablebases are installed (as dennis1382 said)
TordRomstad: Martin: I guess the problem is that Jim doesn't have a CPU with the popcnt instruction.
jhoro: because Tod said SF cannot use them. so this is a big disadvantage in the end games
TordRomstad: This means that he probably can't compile a PGO build with popcnt.
jhoro: i mean Tord
Xaurus: Tord, is there not someone else that can compile it? Dann Corbit? All speedup is goodI think with popcnt SF can gain a few more elo from "nothing"
TordRomstad: If somebody else manages to match Jims compilation skills and compiles with popcnt support, we'll probably include that version in our download.
TordRomstad: By the way, there is a popcnt-binary for Mac OS X.
TordRomstad: ChessBomb: I guess Jim uses ICC under Linux, but I really don't know.
Xaurus: Aha, have you done any comparisons?
Xaurus: jhoro yes that is right, SF has no tablebase code (although I think it's quite strong in the endgame even so)
TordRomstad: jhoro: I actually don't think lacking TBs is a disadvantage in practical endgame play.
Xaurus: ChessBomb, are you Peter?
TordRomstad: Martin: What sort of comparisions? Between popcnt and non-popcnt binaries in OS X, you mean?
Xaurus: Tord, yes exactly
dennis1382: @jhoro: they BOTH use the 3/4/5 men Nalimov TB's
TordRomstad: If I recall correctly, popcnt gives something like 5% extra speed.
TordRomstad: I can do a quick benchmark right now. Wait a minute...
Xaurus: Is that possible to translate to ELO?
TordRomstad: Yes. Wait two minutes.
ChessBomb: @Xaurus, in this case - yes, although this account is used by others too
jhoro: dennis, i think Tord & Xaurus know better than you & i
dennis1382: it is stated on the official site
Xaurus: ChessBomb, thanks for the email - I'll send the admin an email - but my point was the time displayed on the clock's here - it's pretty off for white
ChessBomb: Off by how much?
Xaurus: white has 3:00 minutes
Xaurus: so off by 15 minutes +
Xaurus: black has 2 minutes left right now
Xaurus: if we ignore offset from reading html file
ChessBomb: Are you sure? White should have just gained 15 minutes after move 60
jhoro: dennis, Xaurus runs the match and Tord writes SF. they both said SF cannot use TBs
ChessBomb: IMHO Fritz is wrong in this case, and here we have it correct
Xaurus: Ah yes you're right - CB gui doesn't add it until it's white 61. turn
Xaurus: my fault
Xaurus: I didn't notice it was move 60 already
Xaurus: It adds it after 60 of course but it won't display it until it's white's 61. move
ChessBomb: Aside from that, up to 3 min. errors should be expected probably, because that's the update frequency
ChessBomb: (as you stated on the site)
Xaurus: yes correct
Xaurus: I am impressed, great work on that code
Xaurus: 1-0, nice win for SF
ChessBomb: And white wins, congrats SF
Xaurus: game 28 already started
Aspirador: who declared this a win?
Aspirador: when do you award the win?
TordRomstad: My quick benchmark gave only 3% extra speed for the popcnt binary.
TordRomstad: That's about 3 Elo points.
Xaurus: Thanks Tord! 3% wasn't very much though
Xaurus: Aspirador, the GUI awards the win as you put it
TordRomstad: I expected more. Perhaps a longer benchmark would have given a different result.
