FuryBench

FuryBench Testing Framework

home to Integral PlentyChess Obsidian Caissa Quanticade Serendipity Spaghet Schoenemann Clarity Illumina Pawnocchio Vine Anura

Finished
Anemat	Quanticade	prefetch-better	diff	4.0+0.04	LLR +2.91 (-2.25, +2.89) bounds [0.00, 3.00] elo ELO +10.21 ± 4.43 (+5.79, +14.64) 95% GAMES 6,570 (26.5% 1,742W 49.9% 3,279D 23.6% 1,549L) PENTA [38⁻² 655⁻¹ 1,717⁰ 826⁺¹ 49⁺²]	take 2
Mergen	Illumina	better-improving-condition	diff	8.0+0.08	LLR +2.90 (-2.25, +2.89) bounds [0.00, 5.00] elo ELO +8.06 ± 4.87 (+3.20, +12.93) 95% GAMES 5,562 (22.7% 1,263W 56.9% 3,165D 20.4% 1,134L) PENTA [26⁻² 624⁻¹ 1,350⁰ 757⁺¹ 24⁺²]	Better improving condition
DarkNe	Quanticade	net59	diff	N=5000	LLR +3.03 (-2.25, +2.89) bounds [0.00, 3.00] elo ELO +2.56 ± 2.03 (+0.54, +4.59) 95% GAMES 63,902 (34.0% 21,698W 32.8% 20,977D 33.2% 21,227L) PENTA [2,719⁻² 6,809⁻¹ 12,565⁰ 6,998⁺¹ 2,860⁺²]	Fixed nodes test to even check
Anemat	Quanticade	reduce-memset	diff	4.0+0.04	LLR +2.99 (-2.25, +2.89) bounds [0.00, 3.00] elo ELO +8.46 ± 3.92 (+4.53, +12.38) 95% GAMES 8,220 (25.9% 2,125W 50.7% 4,170D 23.4% 1,925L) PENTA [40⁻² 855⁻¹ 2,129⁰ 1,037⁺¹ 49⁺²]	blah
Anemat	Quanticade	pext-attacks	diff	4.0+0.04	LLR +2.98 (-2.25, +2.89) bounds [0.00, 3.00] elo ELO +9.97 ± 4.36 (+5.61, +14.33) 95% GAMES 7,040 (26.6% 1,874W 49.6% 3,494D 23.8% 1,672L) PENTA [31⁻² 769⁻¹ 1,743⁰ 921⁺¹ 56⁺²]	pext attacks
Witek	Caissa	singular-exact-margin *	diff	10.0+0.10	LLR +2.96 (-2.94, +2.94) bounds [0.00, 2.00] elo ELO +1.48 ± 1.13 (+0.34, +2.61) 95% GAMES 95,014 (23.8% 22,588W 52.9% 50,242D 23.3% 22,184L) PENTA [304⁻² 11,096⁻¹ 24,283⁰ 11,540⁺¹ 284⁺²]
Fury	Integral	nnue_threat_inputs	diff	40.0+0.40	LLR +2.89 (-2.25, +2.89) bounds [0.00, 4.00] elo ELO +24.43 ± 7.77 (+16.68, +32.21) 95% GAMES 1,866 (28.2% 527W 50.5% 943D 21.2% 396L) PENTA [1⁻² 162⁻¹ 477⁰ 291⁺¹ 2⁺²]	Switch default network
Fury	Integral	vaporeon	diff	4.0+0.04	LLR +2.90 (-2.25, +2.89) bounds [0.00, 2.50] elo ELO +3.27 ± 2.10 (+1.17, +5.37) 95% GAMES 26,660 (25.6% 6,831W 49.7% 13,249D 24.7% 6,580L) PENTA [131⁻² 2,721⁻¹ 7,377⁰ 2,968⁺¹ 133⁺²]
Fury	Integral	ti-768-chef	diff	40.0+0.40	LLR +2.91 (-2.25, +2.89) bounds [0.00, 4.00] elo ELO +17.35 ± 6.50 (+10.86, +23.85) 95% GAMES 2,666 (26.9% 717W 51.2% 1,365D 21.9% 584L) PENTA [3⁻² 249⁻¹ 696⁰ 382⁺¹ 3⁺²]
Fury	Integral	ti-768-chef	diff	10.0+0.10	LLR +2.89 (-2.25, +2.89) bounds [0.00, 4.00] elo ELO +5.70 ± 3.58 (+2.12, +9.27) 95% GAMES 9,700 (26.3% 2,555W 49.0% 4,749D 24.7% 2,396L) PENTA [31⁻² 1,089⁻¹ 2,458⁰ 1,234⁺¹ 38⁺²]
Fury	Integral	ti-768-chef	diff	N=20000	LLR +2.94 (-2.25, +2.89) bounds [0.00, 4.00] elo ELO +10.15 ± 5.53 (+4.62, +15.68) 95% GAMES 6,714 (31.7% 2,129W 39.5% 2,652D 28.8% 1,933L) PENTA [158⁻² 729⁻¹ 1,417⁰ 865⁺¹ 188⁺²]
DarkNe	Quanticade	lmp-table	diff	4.0+0.04	LLR +2.92 (-2.25, +2.89) bounds [0.00, 3.00] elo ELO +3.13 ± 2.17 (+0.96, +5.30) 95% GAMES 24,858 (24.7% 6,139W 51.5% 12,804D 23.8% 5,915L) PENTA [75⁻² 2,693⁻¹ 6,711⁰ 2,833⁺¹ 117⁺²]	Source: Move LMP formula into a lookup table
Fury	Integral	ti-768-rfs	diff	10.0+0.10	LLR +2.90 (-2.25, +2.89) bounds [0.00, 2.50] elo ELO +1.97 ± 1.50 (+0.47, +3.47) 95% GAMES 54,256 (25.2% 13,660W 50.2% 27,243D 24.6% 13,353L) PENTA [156⁻² 6,347⁻¹ 13,836⁰ 6,612⁺¹ 177⁺²]
DarkNe	Quanticade	master	diff	10.0+0.10	ELO +71.22 ± 3.66 (+67.58, +74.87) 95% GAMES 10,022 of 10,000 (36.1% 3,619W 48.0% 4,810D 15.9% 1,593L) PENTA [10⁻² 500⁻¹ 2,147⁰ 2,162⁺¹ 192⁺²]	Test against latest release STC
Fury	Integral	ti-768-vf-1	diff	10.0+0.10	LLR +2.91 (-2.25, +2.89) bounds [0.00, 2.50] elo ELO +2.87 ± 1.95 (+0.92, +4.81) 95% GAMES 32,856 (25.6% 8,414W 49.6% 16,299D 24.8% 8,143L) PENTA [113⁻² 3,824⁻¹ 8,281⁰ 4,099⁺¹ 111⁺²]
Fury	Integral	perf-sweep	diff	4.0+0.04	LLR +2.90 (-2.25, +2.89) bounds [0.00, 2.50] elo ELO +4.21 ± 2.45 (+1.77, +6.66) 95% GAMES 19,712 (26.1% 5,144W 49.0% 9,663D 24.9% 4,905L) PENTA [101⁻² 1,971⁻¹ 5,487⁰ 2,182⁺¹ 115⁺²]	death by a thousand papercuts
Fury	Integral	ti_tweaks	diff	40.0+0.40	ELO +95.12 ± 3.14 (+92.00, +98.26) 95% GAMES 10,010 of 10,000 (39.0% 3,899W 48.8% 4,886D 12.2% 1,225L) PENTA [4⁻² 238⁻¹ 1,883⁰ 2,840⁺¹ 40⁺²]
Fury	Integral	ti_tweaks	diff	40.0+0.40	LLR +2.90 (-2.25, +2.89) bounds [0.00, 2.50] elo ELO +16.91 ± 5.11 (+11.81, +22.02) 95% GAMES 4,256 (27.2% 1,159W 50.4% 2,145D 22.4% 952L) PENTA [6⁻² 394⁻¹ 1,116⁰ 611⁺¹ 1⁺²]
Fury	Integral	ti-vf-768	diff	10.0+0.10	LLR +2.92 (-2.25, +2.89) bounds [0.00, 2.50] elo ELO +15.63 ± 5.02 (+10.61, +20.65) 95% GAMES 4,850 (28.4% 1,376W 47.8% 2,316D 23.9% 1,158L) PENTA [13⁻² 470⁻¹ 1,256⁰ 658⁺¹ 28⁺²]
Fury	Integral	ti-768-vf	diff	N=20000	LLR +3.07 (-2.25, +2.89) bounds [0.00, 5.00] elo ELO +17.64 ± 8.18 (+9.48, +25.82) 95% GAMES 3,134 (33.2% 1,039W 38.8% 1,215D 28.1% 880L) PENTA [70⁻² 318⁻¹ 664⁰ 413⁺¹ 102⁺²]
Vast	Clarity	bestmove *	diff	8.0+0.08	LLR +2.95 (-2.25, +2.89) bounds [-3.00, 1.00] elo ELO +4.35 ± 3.68 (+0.66, +8.03) 95% GAMES 9,754 (23.9% 2,332W 53.4% 5,212D 22.7% 2,210L) PENTA [55⁻² 1,104⁻¹ 2,444⁰ 1,212⁺¹ 62⁺²]	Get best move from PV table I expect null moves, which I'll fix later (hopefully later today)
Witek	Caissa	fix-bmi2	diff	5.0+0.05	LLR +2.95 (-2.94, +2.94) bounds [0.00, 2.00] elo ELO +3.01 ± 1.84 (+1.17, +4.85) 95% GAMES 37,274 (24.9% 9,297W 51.0% 19,003D 24.1% 8,974L) PENTA [203⁻² 4,093⁻¹ 9,721⁰ 4,418⁺¹ 202⁺²]	Fix pext/pdep and MulHi64 falling back to software on non-Windows builds Also, with pext working, restore -DUSE_BMI2 in the makefile native probe. No functional change.
Witek	Caissa	avxvnni_no_bmi2	diff	5.0+0.05	LLR +2.95 (-2.94, +2.94) bounds [0.00, 2.00] elo ELO +3.86 ± 2.15 (+1.71, +6.00) 95% GAMES 28,568 (25.1% 7,166W 50.9% 14,553D 24.0% 6,849L) PENTA [148⁻² 3,252⁻¹ 7,222⁰ 3,459⁺¹ 203⁺²]	AVX-VNNI support
Mergen	Illumina	pre-v4-tune	diff	40.0+0.40	LLR +2.89 (-2.25, +2.89) bounds [0.00, 5.00] elo ELO +3.15 ± 2.52 (+0.64, +5.67) 95% GAMES 18,394 (21.0% 3,865W 58.9% 10,831D 20.1% 3,698L) PENTA [36⁻² 2,074⁻¹ 4,817⁰ 2,227⁺¹ 43⁺²]	Tuning
Fury	Integral	ti_tweaks	diff	20.0+0.20	LLR +2.91 (-2.25, +2.89) bounds [0.00, 4.00] elo ELO +12.31 ± 5.43 (+6.88, +17.74) 95% GAMES 3,868 (26.6% 1,029W 50.3% 1,947D 23.1% 892L) PENTA [4⁻² 395⁻¹ 998⁰ 534⁺¹ 3⁺²]

1 2 3 55 56 57