For periodic calculations you probably want to check convergence of your desired property with respect to kpoints.
E.g. if you want to optimize lattice constants + coordinates, increase the kpoint grid until you are happy with the convergence within a certain limit (1% of lattice parameters, 0.01 A for coordinates ??)
But perhaps you can make an example yourself. Build bulk Silicon and see how the band gap is affected by kpoints. Or optimize the lattice of quartz ....
Eugene F Kislyakov, Sorry sir I can not follow your statement since I am a novice who is using DFTB+ for the first time.I request to have it in little more descriptive .
In principle you dont need to use any implementation of kpoints in dftb+, the program will create a big unit cell internally when you dont switch on the periodicity flag. So when you want to calculate the molecular properties of a system just give the coordinates in cartesian and try to move your molecule top the Center of Mass in any visualization program. I hope this helps you