One platform, two mechanical branches
FLEX 9 comprises four models: J100E and J130E in E3.S, J100S and J130S in E1.S, all PCIe 5.0 and aimed at OCP cloud data centres and high-density nodes. Splitting into two branches was not about inflating the model count — firmware, diagnostics and management interfaces are shared. What actually diverges is the mechanical and thermal envelope that carries the drive. All four models are still moving between product definition and EVT, so the form-factor differences below are design-stage engineering judgement and will converge as prototype measurements come in.
Thermals: the question shifts from "does it fit" to "does it hold speed at full load"
A Gen5 x4 link doubles the signalling rate of Gen4 x4, and the baseline power of the controller and PHY rises with it, which makes cooling the first constraint on form factor choice. E1.S is limited in both board area and thickness; EDSFF defines several thickness options for it, and thinner variants leave less room for thermal interface material and heat-spreading structure, so less heat leaves the drive at the same airflow. E3.S offers more board area and internal volume, allowing thicker interface material and a larger spreader to distribute heat. The difference does not show up in peak numbers — it shows up in when thermal throttling engages during sustained sequential transfers. The two branches also cannot share one thermal conclusion: the steady-state temperature rise of the same controller in two different enclosures cannot be extrapolated from one to the other, so thermal verification has to be repeated per form factor.
Power-loss protection: trading hardware energy storage against firmware flush strategy
PLP depends on on-board energy storage, and those components consume both area and height. E1.S limits how many can be placed and where, so the engineering preference is fewer parts, paired with firmware work to reduce the volume of data that must be flushed on power loss and shorten the flush window. E3.S has more structural headroom and can afford a more generous storage layout, buying a larger write buffer and a less aggressive flush timeline. The two branches therefore sit at different points on the same trade curve, while the enterprise feature baseline stays identical: end-to-end protection, T10 DIF/DIX, multiple namespaces, NVMe-MI and TCG OPAL / Secure Boot / Sanitize are one implementation across both form factors.
Carrier, backplane and chassis: the customer is often not choosing a drive
E1.S typically serves dense front-panel slot arrays, yielding more bays per 1U node at the cost of tighter backplane routing density and signal integrity margins, which Gen5 rates squeeze further on connectors and cabling. E3.S is closer to a drop-in path for traditional 2.5-inch bays, requiring smaller changes to airflow and chassis structure while allowing more media and a larger power budget per drive. In real projects the decision is usually dictated by the existing chassis, backplane and fan policy rather than by the drive itself.
Why both must coexist
Compute-heavy nodes care about bays per rack unit and system-level power allocation, where E1.S stacks density more easily within a constrained air path. Storage and AI nodes care about capacity per drive and sustainable bandwidth, where the thermal and power headroom of E3.S suits long high-load runs better. Keeping both on one firmware and management plane — telemetry, online update and rollback, low-latency QoS, hot plug — means operations teams do not maintain two separate procedures; SR-IOV and OCP-specific capabilities are scheduled as P1-phase deliverables. The advice to customers is therefore plain: confirm which slot form factor the chassis and backplane have already committed to, and only then discuss capacity and performance per drive. Projects that reverse that order usually end up reworking at system integration.