PERC vs Smart Array: Understanding RAID Controllers

PERC and Smart Array are the brand names Dell and HPE use for the RAID controllers in their servers. The silicon underneath is shared less often than the similar feature lists suggest. Dell's PERC line has been built on LSI, now Broadcom, MegaRAID controllers for years. HPE's Smart Array SR controllers — P420i, P440ar, P408i-a, P816i-a and their siblings — run Microsemi, now Microchip and before that PMC-Sierra and Adaptec, SmartIOC and SmartROC parts. Only HPE's newer MR series is MegaRAID-based. The differences that matter in production sit on top of that: firmware behaviour, how cache is protected during power loss, what the defaults are, and what happens when you move a set of drives from one machine to another.

Reading the model numbers

Dell's PowerEdge RAID Controller numbering encodes class first and generation second. The leading digit is the class — 3-series entry controllers, 7-series cached performance controllers, 8-series cards with external ports — and the middle digit tracks the server generation. That is why the entry line reads H310, H330, H345 across three platform generations while the performance line reads H710, H730, H740, H745. A trailing P marks the larger-cache variant of the same controller: an H730P carries more write cache than an H730, an H710P more than an H710. Form factor is called out separately: Mini or Mini Mono controllers drop into a dedicated internal connector and consume no PCIe slot, while Adapter versions are standard add-in cards. The HBA330 and later HBA models are not RAID controllers at all.

HPE encodes more in the string. The leading letter is the class: S is chipset-based software RAID, E is an entry controller without cache, P is a cached performance controller. Then comes the port count and generation, then i for internal ports or e for external, and finally the mounting style: -a for a modular AROC that sits in its own dedicated connector, -p for a standup PCIe card, -c for the custom form-factor cards used in blades and Apollo nodes, and -m for mezzanine cards such as the P416ie-m. So a P408i-a is a cached controller with eight internal ports in the modular slot.

One more HPE distinction is easy to miss and changes your entire tooling story: SR controllers run HPE's own Smart Array firmware and are managed with ssacli, while MR controllers, introduced on Gen10 Plus, run Broadcom MegaRAID firmware and are managed with storcli. Scripts written for one will not run against the other.

Platform generation	Dell PERC	HPE Smart Array	Interface
Dell 12G / HPE Gen8	H310, H710, H710P	P420i, P420, P822	6 Gb SAS; PCIe 2.0 on the PERCs, PCIe 3.0 on the Smart Arrays
Dell 13G / HPE Gen9	H330, H730, H730P	P440ar, P440, P840, H240 HBA	12 Gb SAS, PCIe 3.0
Dell 14G / HPE Gen10	H330, H730P, H740P, H840, HBA330	E208i-a, P408i-a, P816i-a	12 Gb SAS, PCIe 3.0
Dell 15G / HPE Gen10 Plus	H345, H355, H745, H755, H755N	MR216i-a, MR416i-a	12 Gb SAS, tri-mode NVMe on some, PCIe 4.0

Cache, backup power, and the write-back cliff

Write-back caching is where most of the performance of a parity array comes from, and it is only safe if the cache contents survive a power cut. Both vendors solve that with flash-backed cache: a supercapacitor holds the controller up long enough to flush DRAM contents into onboard NAND, and the data is replayed at next boot.

Battery-backed cache is the older approach and it has not disappeared from the installed base. Dell's 12G controllers shipped with a lithium pack that degraded over time and needed periodic relearn cycles, during which write caching is disabled. HPE had already moved its cached controllers to flash-backed cache with a capacitor pack by that era, so a Gen8 Smart Array and a 12G PERC of the same vintage do not behave the same way here.

The physical arrangement differs in a way that affects spares planning. Dell pairs each cached PERC with its own supercapacitor module and cable. HPE moved to a chassis-level Smart Storage Battery that powers several controllers from one pack, which means one failed battery can drop write caching on every array in the box at once.

That drop is the failure mode to watch for. When backup power is missing, degraded or still charging after a cold boot, firmware falls back to write-through. Nothing is lost, no alert screams at you beyond a health warning, and random write throughput on a RAID 5 or RAID 6 volume collapses. If a server suddenly feels an order of magnitude slower after maintenance, check cache policy before you check anything else.

HPE also exposes a read/write cache ratio you can tune, letting you weight the cache toward writes on a database volume. Dell exposes the policy as discrete settings instead: Write Back, Force Write Back and Write Through, plus Read Ahead or No Read Ahead. Force Write Back keeps caching enabled even with no working backup power — useful in a lab, reckless in production.

RAID levels, strip size, and scrubbing

The cached models on both sides cover RAID 0, 1, 5, 6, 10, 50 and 60. Entry controllers do not, and this is a routine cause of a rebuilt array on install day: Dell's H310 and H330 have no RAID 6 or 60, the H345 is limited to RAID 0, 1 and 10, and HPE's E208i-a stops at RAID 0, 1, 5 and 10. Check the level you intend to run against the exact controller model before ordering, not against the family.

HPE Smart Array SR controllers additionally offer Advanced Data Mirroring — RAID 1 ADM and RAID 10 ADM — which is a three-way mirror that tolerates two simultaneous drive losses in a mirror set. That option went away when HPE moved to the MegaRAID-based MR controllers, so a Gen9 or Gen10 array using ADM cannot be carried forward unchanged.

Strip size defaults are not the same between vendors, or even consistently the same across controller families within a vendor. Check the value before you create the volume rather than after: changing it later means destroying the virtual disk and rebuilding it. Small strips suit random small-block workloads; large strips suit sequential streaming.

Background scrubbing carries different names for the same necessity. Dell calls it Patrol Read, with a separate Check Consistency task for parity verification. HPE calls it Surface Scan Analysis, alongside parity initialization options. Both are on by default in most firmware, and both should stay on — an unread bad block only announces itself during a rebuild, which is the worst possible time. Rebuild aggressiveness is also configurable: Dell as a percentage of controller effort, HPE as a low/medium/high priority.

On modern high-capacity drives, rebuild windows are long enough that RAID 5 is hard to justify for anything you care about. RAID 6 or RAID 10 is the sane default on the larger SAS and SATA capacities we see in the refurbished server configurations we build.

HBA mode, foreign imports, and encryption

When you want the controller out of the way

ZFS, Storage Spaces Direct, vSAN and Ceph all want raw devices, not virtual disks. A RAID controller sitting in front of them hides SMART data, adds a caching layer the filesystem does not know about, and interferes with the software's own redundancy. Use a true HBA: Dell's HBA330 and later, or HPE's H240 in HBA mode. Several PERC and Smart Array models offer a pass-through or mixed mode that is close enough for many purposes, but firmware behaviour around SMART and device serial numbers varies by revision — verify on the exact firmware before committing a storage design to it.

Moving drives between machines

Array metadata lives on the drives, so a complete set can usually be moved to another server with a controller from the same family and imported. Dell surfaces this as foreign configuration import in the BIOS utility or iDRAC; HPE generally detects and adopts a compatible array on boot. Two rules make this survivable: move the whole set, and keep the slot order. Cross-vendor migration does not work at all — a Smart Array set will not import on a PERC, and the recovery path is restore from backup.

Encryption

Both vendors support self-encrypting drives with controller-managed keys, Dell through local key management and Secure Enterprise Key Manager on the cached PERC models, HPE through Smart Array Secure Encryption. Licensing is not the same on both sides: HPE Secure Encryption is a licensed feature, while on Dell local key management works without one and only SEKM requires an iDRAC Datacenter licence. In either case the key management choice is not reversible without wiping the array. It also gives you instant cryptographic erase at decommissioning time, which is worth having if drives ever leave your building. None of it substitutes for backups: RAID protects against a drive failing, not against a deletion, a ransomware event or a controller writing garbage.

Where to go from here

Controller choice should follow the storage design, not the other way round. If you are running a hypervisor on local disks, a cached PERC or Smart Array with working backup power is the right answer; if you are building a software-defined pool, specify an HBA from the start and save yourself a reconfiguration. Browse the refurbished servers to see which controller generations pair with which chassis, or use Build Your Server to pick the controller, cache size, backup module and drive layout you want and send it to us for a quote.