| Active : 10 Machines / 100 Threads / 61.86 MNPS | ||||
|---|---|---|---|---|
| Priority 0 | ||||
| Caissa Witek | NN: Faster threat delta generation The bitboards a threat lookup needs (occupancy, victim masks per attacker type, and the piece code of every square as four bit planes) are computed once per position. Records are written through uint8_t, which may alias the position, so reading it directly reloaded its bitboards after every record. The changed squares come from the planes. Sliders are looked up only through squares whose occupancy changed: a square occupied in both positions stops the ray in both. speedup.py, MinGW bmi2 with eval-ti-2-120B-permuted embedded, depth 12, 20 rounds: 1,835,108 vs 1,814,096 nps, +1.173% [+0.640%, +1.706%]. No functional change. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> | diff | 10.0+0.10 | ELO 3.72 +- 3.85 (95%) LLR 0.80 (-2.94, 2.94) [0.00, 2.00] GAMES N: 7008 W: 1717 L: 1642 D: 3649 PENTA [3, 734, 1958, 803, 6] |
| Caissa Witek | NN: Faster threat delta generation The bitboards a threat lookup needs (occupancy, victim masks per attacker type, and the piece code of every square as four bit planes) are computed once per position. Records are written through uint8_t, which may alias the position, so reading it directly reloaded its bitboards after every record. The changed squares come from the planes. Sliders are looked up only through squares whose occupancy changed: a square occupied in both positions stops the ray in both. speedup.py, MinGW bmi2 with eval-ti-2-120B-permuted embedded, depth 12, 20 rounds: 1,835,108 vs 1,814,096 nps, +1.173% [+0.640%, +1.706%]. No functional change. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> | frc diff | 10.0+0.10 | ELO 1.32 +- 2.34 (95%) LLR 0.60 (-2.94, 2.94) [0.00, 2.00] GAMES N: 27028 W: 6666 L: 6563 D: 13799 PENTA [280, 3175, 6456, 3368, 235] |
| Quanticade DarkNeutrino | Because im a moron who didnt setup the tune correctly previously. | crash smp diff | 20.0+0.20 | TUNE 134 parameters ITER 5,532 / 25,000 GAMES 44,260 / 200,000 |
| Quanticade DarkNeutrino | net63 Just 1B more data | diff | 60.0+0.60 | ELO -1.99 +- 3.32 (95%) LLR -1.37 (-2.25, 2.89) [0.00, 3.00] GAMES N: 9772 W: 2277 L: 2333 D: 5162 PENTA [3, 1145, 2650, 1081, 7] |
| Finished | ||||
| Caissa Witek | NN: Faster threat delta generation The bitboards a threat lookup needs (occupancy, victim masks per attacker type, and the piece code of every square as four bit planes) are computed once per position. Records are written through uint8_t, which may alias the position, so reading it directly reloaded its bitboards after every record. The changed squares come from the planes. Sliders are looked up only through squares whose occupancy changed: a square occupied in both positions stops the ray in both. speedup.py, MinGW bmi2 with eval-ti-2-120B-permuted embedded, depth 12, 20 rounds: 1,835,108 vs 1,814,096 nps, +1.173% [+0.640%, +1.706%]. No functional change. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> | diff | 60.0+0.60 | ELO 14.68 +- 5.22 (95%) LLR 2.95 (-2.94, 2.94) [0.00, 3.00] GAMES N: 4310 W: 1126 L: 944 D: 2240 PENTA [7, 426, 1112, 598, 12] |
| Caissa Witek | NN: Faster threat delta generation The bitboards a threat lookup needs (occupancy, victim masks per attacker type, and the piece code of every square as four bit planes) are computed once per position. Records are written through uint8_t, which may alias the position, so reading it directly reloaded its bitboards after every record. The changed squares come from the planes. Sliders are looked up only through squares whose occupancy changed: a square occupied in both positions stops the ray in both. speedup.py, MinGW bmi2 with eval-ti-2-120B-permuted embedded, depth 12, 20 rounds: 1,835,108 vs 1,814,096 nps, +1.173% [+0.640%, +1.706%]. No functional change. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> | diff | 10.0+0.10 | ELO 2.65 +- 1.71 (95%) LLR 2.97 (-2.94, 2.94) [0.00, 2.00] GAMES N: 45312 W: 11222 L: 10877 D: 23213 PENTA [241, 5355, 11096, 5746, 218] |
| PlentyChess Yoshie2000 | diff | 8.0+0.08 | ELO -33.71 +- 8.46 (95%) LLR -2.26 (-2.25, 2.89) [0.00, 2.50] GAMES N: 1706 W: 325 L: 490 D: 891 PENTA [9, 285, 426, 128, 5] | |
| Quanticade DarkNeutrino | net63 Just 1B more data | diff | 10.0+0.10 | ELO -1.51 +- 2.67 (95%) LLR -1.85 (-2.25, 2.89) [0.00, 3.00] GAMES N: 16586 W: 3983 L: 4055 D: 8548 PENTA [46, 1975, 4321, 1907, 44] |
| Quanticade DarkNeutrino | net63 Just 1B more data | diff | N=5000 | ELO 7.06 +- 4.05 (95%) LLR 3.10 (-2.25, 2.89) [0.00, 3.00] GAMES N: 15548 W: 5345 L: 5029 D: 5174 PENTA [586, 1665, 3066, 1761, 696] |
| Integral Fury | Decay history malus by moves searched for all quiet histories (#271) | diff | 40.0+0.40 | ELO 48.04 +- 3.28 (95%) GAMES N: 10000/10000 W: 3113 L: 1739 D: 5148 PENTA [2, 584, 2480, 1906, 28] |
| PlentyChess Yoshie2000 | diff | N=20000 | ELO -27.94 +- 8.50 (95%) LLR -2.29 (-2.25, 2.89) [0.00, 2.50] GAMES N: 2418 W: 587 L: 781 D: 1050 PENTA [54, 364, 546, 212, 33] | |
| PlentyChess Yoshie2000 | diff | 8.0+0.08 | ELO -76.65 +- 11.62 (95%) LLR -2.27 (-2.25, 2.89) [0.00, 2.50] GAMES N: 958 W: 142 L: 350 D: 466 PENTA [17, 216, 205, 40, 1] | |
| PlentyChess Yoshie2000 | diff | N=20000 | ELO -132.95 +- 18.54 (95%) LLR -2.51 (-2.25, 2.89) [0.00, 2.50] GAMES N: 852 W: 122 L: 433 D: 297 PENTA [99, 157, 134, 28, 8] | |
| Clarity Vast | new smolnet | diff | 8.0+0.08 | ELO -34.86 +- 10.51 (95%) LLR -2.30 (-2.25, 2.89) [0.00, 3.00] GAMES N: 1650 W: 350 L: 515 D: 785 PENTA [43, 268, 342, 155, 17] |
| Clarity Vast | new smolnet | diff | N=20000 | ELO -37.51 +- 10.93 (95%) LLR -2.33 (-2.25, 2.89) [0.00, 3.00] GAMES N: 1590 W: 303 L: 474 D: 813 PENTA [47, 261, 320, 150, 17] |
| Clarity Vast | progtest (bored) | diff | 8.0+0.08 | ELO 14.96 +- 4.42 (95%) GAMES N: 7504/7500 W: 2000 L: 1677 D: 3827 PENTA [76, 751, 1752, 1120, 53] |
| Caissa Witek | SEE: Fix early exit for quiet moves with a positive threshold | diff | 8.0+0.08 | ELO -0.00 +- 0.82 (95%) LLR 2.98 (-2.94, 2.94) [-2.00, 0.00] GAMES N: 180428 W: 43368 L: 43369 D: 93691 PENTA [656, 21261, 46380, 21262, 655] |
| Quanticade DarkNeutrino | SMP VLTC tune preparing for release time | smp diff | 20.0+0.20 | TUNE 134 parameters ITER 5,000 / 5,000 GAMES 20,002 / 20,000 |
| Caissa Witek | diff | 10.0+0.10 | ELO -1.92 +- 1.81 (95%) LLR -2.95 (-2.94, 2.94) [0.00, 2.00] GAMES N: 37818 W: 8952 L: 9161 D: 19705 PENTA [125, 4651, 9570, 4434, 129] | |