Anyone who has a previous version of Ethereal 13.00+, should have recieved an email from me by now containing a link to download the updated binaries and networks, as well as a comment about updates going foward. If you've not yet gotten yours from andrewgrantethereal@gmail.com, you should check your spam and then reach out to me at either andrewgrantethereal@gmail.com or andrew@grantnet.us. I would prefer to provide any support needed via email, not on talkchess.
Changes since the 13.75 Release:
Code: Select all
14.00: Speedup Accumulator using Tiling (Thanks to Berserk's author Jay)
13.99: Once again enable Smart Syzygy filtering of root moves
13.98: Use one-fourth of set threads to clear the Transposition Table
13.97: Improve Syzygy by reducing look ahead reads on Linux (Thanks Aloril + SF)
13.96: Fix Syzygy issue for wide DTZ tables
13.95: Simplify LMR logic in Ethereal to match that of other engines
13.94: Fix Syzygy issues when probing very long endgames (DTZ > 1000)
13.93: Fix a long standing, non-functional bug when storing null values into the TT
13.92: New Fischer Network: Repeat process using 13.85 Network for Data Generation
13.91: Simplify the History component of LMR during Tactical moves
13.90: Improve speed to report TB Win/Loss significantly
13.89: Default to using Clang and PGO builds
13.88: Increase ProbCut Margin from 80cp to 100cp
13.87: Perform 4x maddubs operations per madd(ones, acc). Big speed up.
13.86: New Standard Network: Repeat process using 13.84 Network for Data Generation
13.85: New Fischer Network: Same data as 13.80 final, but 2x512 L1
13.84: New Standard Network: Same process, but with a 2x512 L1 instead of 2x384
13.83: New Standard Network: Same process, but with a 2x384 L1 instead of 2x256
13.82: New Standard Network: 100m games of nodes=40k, 40x LR drop instead of 10x LR drop
13.81: Accept lower-depth tt-entries if we expect a research may follow (big brain idea)
13.80: New Fishcher Network: 100m games of nodes=20k, first network in a while +85 elo
13.79: New Standard Network: 100m games of nodes=20k, 40x LR drop instead of 10x LR drop
13.78: New Standard Network: 100m games of nodes=5k, 40x LR drop instead of 10x LR drop
13.77: Update some of the internal NNUE data processing tools internally
13.76: Simplify singularity() to simply wrap search() and reuse the MovePickerCode: Select all
Standard Chess using a very drawish book:
ELO | 45.54 +- 1.58 (95%)
CONF | 60.0+0.60s Threads=1 Hash=64MB
GAMES | N: 40000 W: 7091 L: 1878 D: 31031
http://chess.grantnet.us/test/30795/Code: Select all
Fischer Random Chess using a very shallow book:
ELO | 101.30 +- 2.08 (95%)
CONF | 60.0+0.60s Threads=1 Hash=64MB
GAMES | N: 40000 W: 13562 L: 2219 D: 24219
http://chess.grantnet.us/test/30797/
.