Server Drive Compatibility Caddies and Firmware

Four Things Have To Match Not One

Buying a server drive on capacity alone is the most common way a compute order stalls. Capacity is the least constrained of the attributes involved.

Caddies Are the Part Everyone Forgets

A bare drive without a caddy cannot be installed, and caddies are specific to the chassis generation rather than universal. On a node build this is the line that most often turns a delivery into a partial delivery.

The practical rule is that a drive line on a bill of materials is incomplete until it states whether the caddy is included. The same applies to rails and cable management on the chassis itself.

Firmware Level Is Part of the Specification

Enterprise drives carry firmware that the controller and the array read. A drive supplied at a level the estate has not qualified can be rejected, or worse, accepted with behaviour that only shows up during a rebuild.

Where an estate runs a qualified firmware level, that level belongs on the purchase order alongside the part number. It is a specification, not a detail to be resolved on site.

Memory Follows the Same Logic

Memory kits are specified by more than capacity. Rank, voltage, native speed and registration type all have to match what the installed processor and chassis will accept, and a mismatch usually means the node will not post rather than that it runs slowly.

• DIMM type and whether the chassis takes registered or unbuffered modules
• Rank, which affects how many modules a channel will accept
• Native speed, which is capped by the processor rather than the module
• Voltage, particularly on older low voltage platforms

Product pages for HPE memory kits on this site list these attributes where the manufacturer publishes them, which is what a compatibility check needs.

Mixed Batches Behave Differently

Drives bought across several months from several sources can be within specification and still rebuild at different rates or report differently under load. In an array this shows up as inconsistent rebuild times rather than as a failure.

For arrays and for repeatable node builds, allocating from one batch makes the estate predictable. It also makes a future fault easier to diagnose because the population is uniform.

A Node Is Complete or It Is Not

The test of a server order is whether the node can be racked and powered on the day it arrives. That depends on a list that is mostly not the server.

• Rails matched to the rack depth and the chassis generation
• Drive caddies for every drive bay that will be populated
• Controller at the firmware level the estate has qualified
• Power supplies matched to the input feed and the node's draw
• Cable management and the correct power leads for the destination country