Poor results with NoahMP versus Noah

Hi,
I currently use Noah and have a small (-1c) cold bias - I read on this forum that NoahMP can improve this situation.
Unfortunately just from toggling Noah to NoahMP I get catastrophically wrong results, -3c cold bias and +5c dewpoint bias. This was surprising as I understood they should behave similarly.
Is there some steps needed other than sf_surface_physics needed to get good results with NoahMP when coming from Noah?
I use 21-class land use and GFS for input.

Would you please upload your namelist.input for me to take a look? Also please let me know which version of WRF you are using. Thanks.

Hi,
See linked the namelist.input:
https://pastebin.com/ZP8BzpF0

There are too many factors that may lead to the poor performance of NoahMP
(1) For the 12km domain, please turn on cumulus scheme
(2) Please turn off slope_rad, mosaic_lu, sf_surface_mosaic, etc., and also turn off adaptive time step. This is to simplify the options and help us to figure out possible problems in NoahMP
(3) Please try rrtmg radiation and see whether you can find the similar model behavior.

Hi Ming,
Thanks for getting back to me.
1) Cumulus is disabled already
2) Confirmed disabled slope_rad, mosaic_lu, sf_surface_mosaic and adaptive timestep
3) I am using 24, which I understand is "fast" RRTMG - I have changed to 4, plain RRTMG.
Unfortunately the results were slightly better but still very similar, still +3c dewpoint bias.

Would you please clarify how you verify the results of Noah and NoahMP?

I analysed both domains against observations from GDAS, comparing the average/median/stddev error.
But I can also just load up all kinds of subsequent parameters and see subsequent catastrophically wrong results, like PBLH being nearly zero across large areas at the height of summer.
Surface soil moisture also appears to be very anomalous.