Hier ein post des Entwicklers:

<@!510939995851325471> thank you for the test of net 610024, this is extremely encouraging and I hope this big progress over J94 can be confirmed in other tests. I looked over the results and games and don't see any problems with the test setup. My only comment would be that possibly (but not likely) the results could be different when using timed controls rather than fixed nodes; potentially the 610024 builds tree shapes which are less efficient to be processed by the engine and would produce lower nps.

Oder: 610034

http://zz4032.bplaced.net/lc0/charts.html

Lothar