The memory saving of diskann is actually based on two thing,

1. PQ you use -> we recommend to use 1/16 or 1/32 PQ
2. cache size generally speaking diskANN can help you to save 3-10x memories. to learn more about diskANN, check https://zilliz.com/learn/DiskANN-and-the-Vamana-Algorithm

For 10M 3072dim data, a rough guess is 64GB diskANN is good enough.

Another suggestion is to try our managed service zilliz cloud, we offered capacity instance which use a index sharing similar idea of diskANN.