FGRL rating list - 60 min + 15 sec
Stockfish 12 (+51 to Stockfish 11)
Igel 2.7.0 NNUE (+157 to Igel 2.5.0)
http://www.fastgm.de
FGRL rating list, 60 min + 15 sec - 1. Stockfish 12, 11. Igel 2.7.0 NNUE
Moderators: hgm, Rebel, chrisw
-
- Posts: 818
- Joined: Mon Aug 19, 2013 6:57 pm
-
- Posts: 3291
- Joined: Wed Mar 08, 2006 8:15 pm
Re: FGRL rating list, 60 min + 15 sec - 1. Stockfish 12, 11. Igel 2.7.0 NNUE
No scaling problems for SF 12 as someone expected.
Jouni
-
- Posts: 671
- Joined: Sun Jan 26, 2020 10:38 pm
- Location: Turkey
- Full name: Mehmet Karaman
Re: FGRL rating list, 60 min + 15 sec - 1. Stockfish 12, 11. Igel 2.7.0 NNUE
Yes, there ara scaling problems. Look at 10 min+ 6 sec rating list. Stockfish 12 vs Stockfish 11 elo difference is +75 elo at this rating list. But at 60 min + 15 sec rating list Stockfish 12 vs Stockfish 11 elo difference is +55 elo
-
- Posts: 550
- Joined: Tue Nov 19, 2019 8:48 pm
- Full name: Alayan Feh
Re: FGRL rating list, 60 min + 15 sec - 1. Stockfish 12, 11. Igel 2.7.0 NNUE
Elo compression at longer TCs makes comparison harder.
If I just pick SF11 vs SF9, +66 elo at 10m+6s and +65 elo at 60m+15s, it would seem SF11 gains over previous versions hold up better with higher TCs, but once you consider the error bars from small sample size, the trend isn't significant enough.
Time-to-elo testing would be the proper way to go - compare time ratio needed to achieve a given strength
If I just pick SF11 vs SF9, +66 elo at 10m+6s and +65 elo at 60m+15s, it would seem SF11 gains over previous versions hold up better with higher TCs, but once you consider the error bars from small sample size, the trend isn't significant enough.
Time-to-elo testing would be the proper way to go - compare time ratio needed to achieve a given strength