i sat in a roadmap a while ago for primera, that plus some tidbits on the net:
https://blocksandfiles.com/2019/06/20/hpe-primera-hardware-configurations/i would say the big jump is from the 630->650.
630 has one asic, vs 650-670 have 4 asics per node. their is a sizer tool vars/HPE have that show this difference in performance.
i personally fought pricing for a 2 node 650's over the 630. i did not see the value of a A630, the lack of spinning FC at the moment puts the A630 in an extremely narrow use case.
the size limits seem a bit odd per model. they also hint at the limit of iops per model, where adding more ssd's will not add performance, just capacity (node bound/node locked). so at some point its worth jumping to the next model. if you look at their pricing, the nodes and cages are an inconsequential price compared to the price of each SSD. they seem to have skewed their SKU's
depending on # of SSD's you want from the A models (independent of the SSD density/size), it would move you up the model line to get performance out of the node models and their configured SSD's.