DISCLAIMER: This is just my opinion of the way things should be based on my knowledge of statistics, not based on any "official" document or interpretation.
I don't think that the deciding factor for STDEV (sample standard deviation) vs STDEVP (population standard deviation) is Cpk vs Ppk, or subgroups vs individuals. It is (as the name implies) an issue of samples vs entire populations.
"The" standard deviation to characterize a group is STDEVP.
If you draw a sample from a group and calculate STDEVP, you tend to get an answer that is smaller than the STDEVP of the entire group. To fix this, a "fudge factor" is added -- you divide by a slightly smaller number ( (n-1)^0.5 instead of n^0.5 ) which makes the answer bigger by an appropriate amount.
STDEV for a subgroup does a good job of estimating the true STDEVP of the entire group.
If you make 30 parts and use all 30 for a Ppk study, then STDEVP would be appropriate.
If you make 3000 parts and use just 30 for a Ppk study, then STDEV would be appropriate.
IF you make 4 parts at a time and use all 4 as a subgroup, then STDEVP would be appropriate.
IF you make 400 parts at a time and use just 4 as a subgroup, then STDEV would be appropriate.
Practically speaking, it doesn't make a lot of difference, espicially for Ppk. If you have 100 parts in the study, then you will be off by 0.5% if you use the "wrong" standard deviation. Ppk = 1.99 vs Ppk = 2.00 -- big deal.
For Cpk, the difference would be more because the size of the group is smaller -- typically 3-6 in a subgroup. Even here, it ould only be 5-15% difference.
Tim F