Chess2u
Would you like to react to this message? Create an account in a few clicks or log in to continue.

Chess2uLog in

descriptionstockfish chess engine with lazy eval Emptystockfish chess engine with lazy eval

more_horiz
I rewrote the eval section of most recent Stockfish source code to make it faster. This is the fastest version in nodes per section that I can find. Even without an intel compiler to compile the source code, it is still faster. The easiest way to tell the speed is to double click on the .exe file then type bench and compared nodes per second with any other version.

I hope someone can test this version and post the results. Win or lose, it is fun playing with the source code. Out of 100, 15" second games, I won like 50 more games than the most recent version of this engine. I did not test it at longer time frames as I am only interested in bullet chess type games.

This is not a virus but an honest attempt to improve the speed of my favorite chess engine. This is the 64 bit engine not the 32 bit. bigsmile


mediafire.com ?2iossfbtul1ofud

descriptionstockfish chess engine with lazy eval EmptyRe: stockfish chess engine with lazy eval

more_horiz
Thanks, what about Stockfish 64bit (not-popcnt) ?

descriptionstockfish chess engine with lazy eval EmptyStockfish with lazy eval

more_horiz
Can you compile a 32bit version pls..
@wasp90210

descriptionstockfish chess engine with lazy eval EmptyRe: stockfish chess engine with lazy eval

more_horiz
Source so I can compile it?

descriptionstockfish chess engine with lazy eval EmptyRe: stockfish chess engine with lazy eval

more_horiz
5000kn/s faster

descriptionstockfish chess engine with lazy eval EmptyBenchmark on my 3960X at 4800 Mhz

more_horiz
121204Qi
2856
5149248
1802957

SF2.3.1.121912
2856
5149248
1802957

271212
2107
4918443
2334334

251212
2326
5671638
2438365

Seems you're compile (latest score) is fastest indeed on benchmark...

Very curious how fast it would be if special Intel Compiler (version 13) would be used.

descriptionstockfish chess engine with lazy eval Emptystockfish from Github

more_horiz
It would appear, as I indicated earlier, that there was a major regression with the newest versions of the stockfish engine. As such, that would certainly explain the +9 elo that I posted on here. I am currently testing the lazy eval engine with the new and fixed engine that was just released from the Github. Nodes per second appear much faster now that the regression has been fixed. I will post results, once I get to 1,000 or 2,000 games. rolleyes

descriptionstockfish chess engine with lazy eval EmptyStockfish w/ L.E.

more_horiz
As you can see, Stockfish apparently corrected their mistake and fixed the regression. As such, I downloaded the most current code, inserted my own code into it and then played it against the most recent engine uploaded by Jack Sparrow, 12/27/2012, on this site. As shown below, nodes per second are much closer, my engine w/ code = 3680885, while the factory version uploaded by Jack Sparrow = 3647984, which is a net gain of only 33,000 nodes per sec. Consequently, the net gain is only +6 ELO after 2,716 games, 604-504-1608.

Code:

Games Completed = 2716 of 50000 (Avg game length = 13.312 sec)
Settings = RR/32MB/500ms+100ms/M 1000cp for 12 moves, D 150 moves/
Time = 19297 sec elapsed, 335957 sec remaining
 1.  50                         1408.0/2716   604-504-1608     (L: m=57 t=0 i=0 a=447)   (D: r=1344 i=194 f=35 s=4 a=31)   (tpm=97.2 d=15.20 nps=3680885)
 2.  latest commit               1308.0/2716   504-604-1608     (L: m=80 t=0 i=0 a=524)   (D: r=1344 i=194 f=35 s=4 a=31)   (tpm=97.3 d=15.17 nps=3647984)

descriptionstockfish chess engine with lazy eval EmptyRe: stockfish chess engine with lazy eval

more_horiz
@wasp90210 wrote:
As you can see, Stockfish apparently corrected their mistake and fixed the regression. As such, I downloaded the most current code, inserted my own code into it and then played it against the most recent engine uploaded by Jack Sparrow, 12/27/2012, on this site. As shown below, nodes per second are much closer, my engine w/ code = 3680885, while the factory version uploaded by Jack Sparrow = 3647984, which is a net gain of only 33,000 nodes per sec. Consequently, the net gain is only +6 ELO after 2,716 games, 604-504-1608.

Code:

Games Completed = 2716 of 50000 (Avg game length = 13.312 sec)
Settings = RR/32MB/500ms+100ms/M 1000cp for 12 moves, D 150 moves/
Time = 19297 sec elapsed, 335957 sec remaining
 1.  50                         1408.0/2716   604-504-1608     (L: m=57 t=0 i=0 a=447)   (D: r=1344 i=194 f=35 s=4 a=31)   (tpm=97.2 d=15.20 nps=3680885)
 2.  latest commit               1308.0/2716   504-604-1608     (L: m=80 t=0 i=0 a=524)   (D: r=1344 i=194 f=35 s=4 a=31)   (tpm=97.3 d=15.17 nps=3647984)

+6 ELO? No,
Elo difference : +13 Elo
Winning percentage : 51.84%

(by ELO Calculator)
https://www.chess2u.com/h2-chess-elo-rating-calculator

descriptionstockfish chess engine with lazy eval EmptyRe: stockfish chess engine with lazy eval

more_horiz
32bit version please!? sad

descriptionstockfish chess engine with lazy eval Emptyelo calculator

more_horiz
Thanks for the information about the ELO calculator. I had been using bayeselo which computes it much lower. I will use this one in the future if you believe it to be more accurate.

Thanks again for your input.

In regards to the 32 engine compile request, I usually don't compile in 32 bit and will have to study up on it first.

descriptionstockfish chess engine with lazy eval EmptyRe: stockfish chess engine with lazy eval

more_horiz
Looking forward to trying it out. Thanks!

descriptionstockfish chess engine with lazy eval EmptyRe: stockfish chess engine with lazy eval

more_horiz
Nicolas wrote:
@wasp90210 wrote:
As you can see, Stockfish apparently corrected their mistake and fixed the regression. As such, I downloaded the most current code, inserted my own code into it and then played it against the most recent engine uploaded by Jack Sparrow, 12/27/2012, on this site. As shown below, nodes per second are much closer, my engine w/ code = 3680885, while the factory version uploaded by Jack Sparrow = 3647984, which is a net gain of only 33,000 nodes per sec. Consequently, the net gain is only +6 ELO after 2,716 games, 604-504-1608.

Code:

Games Completed = 2716 of 50000 (Avg game length = 13.312 sec)
Settings = RR/32MB/500ms+100ms/M 1000cp for 12 moves, D 150 moves/
Time = 19297 sec elapsed, 335957 sec remaining
 1.  50                         1408.0/2716   604-504-1608     (L: m=57 t=0 i=0 a=447)   (D: r=1344 i=194 f=35 s=4 a=31)   (tpm=97.2 d=15.20 nps=3680885)
 2.  latest commit               1308.0/2716   504-604-1608     (L: m=80 t=0 i=0 a=524)   (D: r=1344 i=194 f=35 s=4 a=31)   (tpm=97.3 d=15.17 nps=3647984)

+6 ELO? No,
Elo difference : +13 Elo
Winning percentage : 51.84%

(by ELO Calculator)
https://www.chess2u.com/h2-chess-elo-rating-calculator


can you tell me about which version it is stockfish? +13 elo this thx
bigsmile

descriptionstockfish chess engine with lazy eval EmptyRe: stockfish chess engine with lazy eval

more_horiz
I wrote some code for the evalution.cpp file and search.cpp file of the most recent version of Stockfish Chess Engine. It did very well in fast game testing, 13 second games, and I forwarded my code to Marco Costalba, one of the creators of the Stockfish Engine for his input. He suggested I test my new code at longer time parameters to see if it still comes out on top. So far, I am 1,000 games into my testing and am up by 30+ games. I will let the engine test for 5 to 8 thousand games and then see if he will incorporate my code into his program.

descriptionstockfish chess engine with lazy eval EmptyRe: stockfish chess engine with lazy eval

more_horiz
Permissions in this forum:
You cannot reply to topics in this forum