I can't speak much for Grytics, but for larger groups I wouldn't definitely try to use a stratified sampling method using the biggest sample you can create.
The bigger the same, the more accurate you will be. Make sure it's of active members though not total members. The challenge is likely to be how you randomly process these people. 100+ people is a good standard to begin with, if you can't do that, use 50. But, honestly, the bigger the sample the more accurate the results.