How close in strength is ScorpioNN to Stoofvlees II a10 ?

[Chessqueen]

I tested Scorpio2.9.1 maddex and Stoofvlees a10.
a11 at the Moment has only 200 games. Results later.
TCEC hardware is not optimal for Stoofvlees. Here 5knps on a Gtx 1650. TCEC : 11 knps ?

I just saw the last two games of Stoofvlees II a 11 and its was hitting 12.6 knps which is twice more than with a GTX 1650, to me that is great, but Scorpion was hitting three times more, closer to 37 knps

[Werner]

Here is the Output of scorpio with my Card:
# nodes = 399689 <0% qnodes> time = 9060ms nps = 44115 eps = 0 nneps = 5183
# Tree: nodes = 64332 depth = 19 pps = 5172 visits = 46703
# qsearch_calls = 0 search_calls = 0
Daniel says nps does not mean much, I shall compare nneps. But TCEC shows the nps I think and I saw here 100-150 knps.
I thought TCEC uses much better grafic cards - so only 2x to gtx 1650 seems to be not great ??

[Daniel Shawul]

Hi Werner,

Yes, nps is meaningless as it counts also nodes where NN evaluation is not done.
TCEC shows the nneps/pps value which are almost the same.The difference between pps and nneps
is that pps includes cached nodes counts so pps could be 2-3x greater when there is a ponder hit.
I think it gets on average 70knps and in endgames closer to 150knps there.

Daniel

[Chessqueen]

Hi Werner,

Yes, nps is meaningless as it counts also nodes where NN evaluation is not done.
TCEC shows the nneps/pps value which are almost the same.The difference between pps and nneps
is that pps includes cached nodes counts so pps could be 2-3x greater when there is a ponder hit.
I think it gets on average 70knps and in endgames closer to 150knps there.

Daniel

Are you working on a newer version of ScorpioNN?, since it will have very tough competitors when it reach Division P against Houdini, Komodo, Alliestein, and mainly versus Stockfisk, and LCZero, In TCEC

PS: How does the INT8 works ?