Why partition at all
Splitting a drive into namespaces usually serves one of three needs: logical isolation between tenants or workloads; separating data that requires different LBA formats, such as protection-information-enabled versus plain; or tiering metadata, journals and user data so each can get its own I/O policy. Attachment governs which controller sees which namespace. On dual-port models such as the TITAN 7 J760D/J780D, both controllers can attach the same namespace, which makes format and state consistency across the two paths a first-order concern.
The block format follows you forever
512-byte logical blocks keep older software stacks happy but each block consumes a mapping-table entry, so metadata overhead is markedly higher at equal capacity. 4K aligns with NAND pages and host pages and is the enterprise default. Carrying T10 protection information calls for 4K plus 64 bytes of metadata. Changing the block format requires a Format, because the mapping granularity itself changes — this is not an online adjustment.
Erase semantics and how long they take
The secure-erase setting in the Format command determines how heavy the operation is: rebuild the format only, erase user data, or perform a cryptographic erase. On a self-encrypting drive, cryptographic erase merely destroys the key and completes almost instantly; a physical erase must walk every block, which is substantial at a 15.36TB capacity point. The controller reports busy throughout, so the host timeout has to be sized accordingly or the drive is misread as unresponsive.
Over-provisioning is pooled drive-wide
Namespace capacity fields are a logical view; physical blocks and spare capacity are still scheduled from one drive-wide pool. Creating many small namespaces has a non-obvious consequence: writes within each namespace struggle to fill a complete superblock, reclaim efficiency drops, and effective over-provisioning erodes. The capacity/endurance trade already shows in the SKU list — J750 at 1.92/3.84/7.68/15.36TB with 1 DWPD planned, J770 at 1.6/3.2/6.4/12.8TB with 3 DWPD; the difference is spare area.
Format is not Sanitize
The two get conflated. Format acts on a single namespace, is fine-grained, and suits routine reconfiguration. Sanitize is drive-wide, spans all namespaces, and once started must run to a completed state — a power cycle mid-operation resumes it rather than cancelling it. Decommissioning and RMA flows belong to the latter. Both sit with NVMe-MI out-of-band management and TCG OPAL in the management-plane portion of JH Semiconductor's enterprise feature baseline.