Komodo Dragon T32 +26 +92/=890/-18 53.70% 537.0/1000
Komodo Dragon T16 -26 +18/=890/-92 46.30% 463.0/1000
Engine Depth Time Games Moves Average MIDG EARLY ENDG LATE
---------------------------------------------------------------------------------------------------
Komodo Dragon T32 25.58 31:50:56 1000 107211 1.07 24.58 | 26.05 | 27.20 | 25.46
Komodo Dragon T16 24.74 31:48:58 1000 107202 1.07 23.66 | 24.81 | 26.23 | 24.76
Legend
Depth : Overall average search depth
Time : Total time engine used
Moves : Total moves engine played
Average : Average time per move in centi-seconds
MIDG : Average search depth during the opening and middle game
EARLY : Average search depth during the middle game without Queens, the so called early end game
ENDG : Average search depth during the end game
LATE : Average search depth during the late end game
mar wrote: ↑Tue Mar 30, 2021 9:24 am
so that pretty much debunks the myth that NNUE-based engines don't scale with lazy smp (assuming that's what Komodo Dragon uses)
This debunks absolutely nothing. There is no baseline to compare. If going from 1 to 2 threads one gains 70Elo but going from 16 to 32 gains only 25Elo talking about good scaling is ridiculous. Or if with classic based eval one gains 40Elo going from 16 to 32 threads and with NNUE eval only 25Elo, again we can't just claim all is great.
Milos wrote: ↑Tue Mar 30, 2021 9:44 am
This debunks absolutely nothing. There is no baseline to compare. If going from 1 to 2 threads one gains 70Elo but going from 16 to 32 gains only 25Elo talking about good scaling is ridiculous. Or if with classic based eval one gains 40Elo going from 16 to 32 threads and with NNUE eval only 25Elo, again we can't just claim all is great.
+25 for 32 vs 16 is pretty good, actually. try this with YBW, then come back
