Unknown Workload action
Active : 2 Machines / 47 Threads / 32.43 MNPS
Priority 0
VshcheIgel020217c0.frc *diff10.0+0.10LLR: 2.56 (-2.94, 2.94) [0.00, 2.00]
Games: 13260 W: 3471 L: 3192 D: 6597
Ptnml(0-2): 104, 1563, 3119, 1638, 206
FRC regression test
VshcheIgel6b953e78.standarddiff10.0+0.10LLR: 1.73 (-2.94, 2.94) [0.00, 2.00]
Games: 6406 W: 1663 L: 1495 D: 3248
Ptnml(0-2): 27, 699, 1577, 879, 21
Check current vs past network - STD
MhouppStashchp_quadratic_threshdiff40.0+0.40LLR: 0.89 (-2.94, 2.94) [0.00, 5.00]
Games: 9234 W: 2179 L: 2110 D: 4945
Ptnml(0-2): 45, 1094, 2295, 1113, 70
Use a quadratic scaling threshold for Continuation History Pruning
JhonnoBerserkexp/h1536diff60.0+0.60Elo: 10.07 +- 21.13 (95%) [N=40000]
Games: 276 W: 66 L: 58 D: 152
Ptnml(0-2): 0, 32, 66, 40, 0
0033 v 0032: N_HIDDEN 1536 vs 1024, both on ob-40847-40884. exp/h1536 450a074 vs net-rearch 2994f74, NNZ-sorted nets. LTC
JhonnoBerserkexp/h1536diff10.0+0.10Elo: 0.00 +- 0.00 (95%) [N=40000]
Games: 0 W: 0 L: 0 D: 0
Ptnml(0-2): 0, 0, 0, 0, 0
0033 v 0032: N_HIDDEN 1536 vs 1024, both on ob-40847-40884. exp/h1536 450a074 vs net-rearch 2994f74, NNZ-sorted nets. STC
VshcheIgel6b953e78.standard *diff10.0+0.10LLR: -1.80 (-2.94, 2.94) [0.00, 2.00]
Games: 13330 W: 3273 L: 3432 D: 6625
Ptnml(0-2): 188, 1639, 3096, 1628, 114
Check current vs past network - FRC
Priority -5
TerjeWeissConnectivity2diff60.0+0.60LLR: -0.23 (-2.94, 2.94) [0.00, 3.00]
Games: 64480 W: 16618 L: 16482 D: 31380
Ptnml(0-2): 438, 7600, 16013, 7766, 423
Finished
VshcheIgelmasterdiff60.0+0.60Elo: 64.26 +- 6.17 (95%) [N=40000]
Games: 2942 W: 993 L: 455 D: 1494
Ptnml(0-2): 4, 137, 657, 663, 10
3.7.13 vs prev. GA (3.7.0) [STD] [LTC]
VshcheIgel6b953e78.standard *diff10.0+0.10LLR: -2.98 (-2.94, 2.94) [-3.00, 1.00]
Games: 16160 W: 4007 L: 4183 D: 7970
Ptnml(0-2): 209, 2009, 3746, 1981, 135
Check current vs past network - FRC
JhonnoBerserk0032-1000s.nndiff10.0+0.10Elo: 1.16 +- 1.85 (95%) [N=40000]
Games: 37992 W: 9643 L: 9516 D: 18833
Ptnml(0-2): 142, 4654, 9303, 4729, 168
0032: 0026 recipe on ob-40847-40884 (ob7 + 40884). Both sides net-rearch 2994f74, NNZ-sorted nets.
VshcheIgelmaster *diff10.0+0.10Elo: 62.88 +- 5.08 (95%) [N=40000]
Games: 5854 W: 1989 L: 941 D: 2924
Ptnml(0-2): 21, 352, 1249, 1168, 137
3.7.13 vs prev. GA (3.7.0) [FRC] [STC] - FRC NET
VshcheIgelmasterdiff10.0+0.10Elo: 62.51 +- 3.58 (95%) [N=40000]
Games: 10000 W: 3414 L: 1634 D: 4952
Ptnml(0-2): 22, 547, 2163, 2165, 103
3.7.13 vs prev. GA (3.7.0) [STD] [STC]
VshcheIgelmaster *diff10.0+0.10Elo: 59.38 +- 18.55 (95%) [N=40000]
Games: 514 W: 184 L: 97 D: 233
Ptnml(0-2): 2, 38, 105, 95, 17
3.7.13 vs prev. GA (3.7.0) [FRC] [STC]
VshcheIgelnmpvsdiff60.0+0.60LLR: 2.96 (-2.94, 2.94) [0.00, 2.00]
Games: 43884 W: 10840 L: 10525 D: 22519
Ptnml(0-2): 22, 5042, 11511, 5333, 34
3.7.13: null move pruning verification search
JhonnoBerserk[Stockfish] a8a4911bf7ff48b607994b3640faaee8fe406113diffN=5000 *Generated 16667868/16666667 Games
Elo: -20.32 +- 0.15 (95%) [N=16666667]
Games: 16667868 W: 6336938 L: 7310876 D: 3020054
JhonnoBerserk0034-h256-3b.nndiff60.0+0.60Elo: 4.20 +- 1.74 (95%) [N=40000]
Games: 40002 W: 10002 L: 9519 D: 20481
Ptnml(0-2): 88, 4601, 10154, 5056, 102
0034 v 0033: h256 data scaling, 3B (all ob7) vs 2B (4 shards) LTC
JhonnoBerserk0033-h256-2b.nndiff60.0+0.60Elo: 4.95 +- 1.73 (95%) [N=40000]
Games: 40004 W: 10122 L: 9552 D: 20330
Ptnml(0-2): 85, 4536, 10209, 5068, 104
0033 v 0032: h256 data scaling, 2B (4 shards) vs 1B (2 shards) LTC
JhonnoBerserk0033-h256-2b.nndiff10.0+0.10Elo: 5.81 +- 1.87 (95%) [N=40000]
Games: 40036 W: 10757 L: 10087 D: 19192
Ptnml(0-2): 249, 4645, 9650, 5135, 339
0033 v 0032: h256 data scaling, 2B (4 shards) vs 1B (2 shards) STC
JhonnoBerserk0034-h256-3b.nndiff10.0+0.10Elo: 2.56 +- 1.89 (95%) [N=40000]
Games: 40114 W: 10591 L: 10296 D: 19227
Ptnml(0-2): 296, 4812, 9588, 5023, 338
0034 v 0033: h256 data scaling, 3B (all ob7) vs 2B (4 shards) STC
VshcheIgelnmpvsdiff10.0+0.10LLR: 2.99 (-2.94, 2.94) [0.00, 2.00]
Games: 60858 W: 15174 L: 14822 D: 30862
Ptnml(0-2): 152, 7009, 15777, 7317, 174
3.7.13: null move pruning verification search
MhouppStashchp_quadratic_threshdiff8.0+0.08LLR: 2.95 (-2.94, 2.94) [0.00, 5.00]
Games: 14994 W: 3727 L: 3547 D: 7720
Ptnml(0-2): 153, 1756, 3526, 1882, 180
Use a quadratic scaling threshold for Continuation History Pruning
MhouppStashlimit_ext_by_depthdiff8.0+0.08LLR: -3.21 (-2.94, 2.94) [0.00, 5.00]
Games: 11340 W: 2695 L: 2776 D: 5869
Ptnml(0-2): 124, 1409, 2682, 1334, 121
Simplify extension limit to only use plies over depth ratio
VshcheIgelsnmpdiff10.0+0.10LLR: -2.95 (-2.94, 2.94) [0.00, 2.00]
Games: 160936 W: 40065 L: 40091 D: 80780
Ptnml(0-2): 546, 19670, 40098, 19572, 582
0.0.13: static null move pruning
MhouppStashlmr_ttpv_v4diff8.0+0.08LLR: -3.02 (-2.94, 2.94) [0.00, 5.00]
Games: 12086 W: 2988 L: 3059 D: 6039
Ptnml(0-2): 141, 1503, 2813, 1458, 128
Decrease reductions for nodes with a TT entry coming from the PV
MhouppStashnew_depth_with_ext_v2diff8.0+0.08LLR: -2.98 (-2.94, 2.94) [0.00, 5.00]
Games: 28734 W: 7063 L: 7063 D: 14608
Ptnml(0-2): 320, 3516, 6718, 3470, 343
Adjust new_depth with extensions before LMR
VshcheIgelprobcut_ttdiff60.0+0.60LLR: 2.95 (-2.94, 2.94) [0.00, 2.00]
Games: 70284 W: 17175 L: 16825 D: 36284
Ptnml(0-2): 41, 8070, 18570, 8420, 41
3.7.12: store probcut result in tt (LTC)
MhouppStashcompact_atomic_ttdiff8.0+0.08LLR: -3.09 (-2.94, 2.94) [0.00, 5.00]
Games: 37474 W: 9237 L: 9205 D: 19032
Ptnml(0-2): 476, 4539, 8646, 4629, 447
Make the TT entries smaller and atomic Reduce the size of TT entries from 16 to 12 bytes by only storing the low 32 bits of the board key in them, as we use the high bits for indexing the table. This allows us to store an extra entry per TT cluster and still have them fit in a single cache line on x86_64 and ARM. We also use this change to switch to atomic accesses for TT entry fields, as on some target 32-bit archs we might have had an issue with non-lock-free 64-bit atomic values previously.
MhouppStashcompact_atomic_ttdiff8.0+0.08LLR: -2.95 (-2.94, 2.94) [0.00, 5.00]
Games: 7106 W: 1698 L: 1788 D: 3620
Ptnml(0-2): 104, 870, 1668, 834, 77
Make the TT entries smaller and atomic Reduce the size of TT entries from 16 to 12 bytes by only storing the low 32 bits of the board key in them, as we use the high bits for indexing the table. This allows us to store an extra entry per TT cluster and still have them fit in a single cache line on x86_64 and ARM. We also use this change to switch to atomic accesses for TT entry fields, as on some target 32-bit archs we might have had an issue with non-lock-free 64-bit atomic values previously.
JhonnoBerserk0031-h256-7k5.nndiff60.0+0.60Elo: 2.83 +- 1.85 (95%) [N=40000]
Games: 40002 W: 10373 L: 10047 D: 19582
Ptnml(0-2): 260, 4667, 9835, 4965, 274
0031 v 0030 LTC: h256, 7.5k-node labels (40855) vs 5k (40847)
VshcheIgelmasterdiff10.0+0.10LLR: 0.34 (-2.94, 2.94) [0.00, 2.00]
Games: 1258 W: 320 L: 287 D: 651
Ptnml(0-2): 3, 143, 307, 170, 6
3.7.11: check Hash 8 Mb with Hash 128 Mb
JhonnoBerserk0031-h256-7k5.nndiff10.0+0.10Elo: 1.02 +- 2.02 (95%) [N=40000]
Games: 40094 W: 11173 L: 11055 D: 17866
Ptnml(0-2): 549, 4826, 9204, 4894, 574
0031 v 0030 STC: h256, 7.5k-node labels (40855) vs 5k (40847)
VshcheIgelprobcut_ttdiff10.0+0.10LLR: 2.96 (-2.94, 2.94) [0.00, 2.00]
Games: 80078 W: 19981 L: 19596 D: 40501
Ptnml(0-2): 235, 9557, 20088, 9906, 253
3.7.12: store probcut result in tt (STC)