Every refurbished HPE or Dell server arrives with a storage controller, and the mode you run it in decides how fast your writes are, which software you can run safely and whether your data can ever leave that server. Here is how Smart Array, PERC and plain HBA mode compare.
The short answer
Hardware RAID with a healthy cache for Windows, Hyper-V, VMware and databases. HBA or IT mode for ZFS, Ceph, vSAN and Storage Spaces Direct. Never RAID 0 per disk.
Compare the three ↓HPE
Hardware RAID
HPE Smart Array
P440ar · P408i-a · P816i-a · SR932i-p · 2 to 8 GB flash-backed cache
Choose it if
- You run ProLiant Gen9, Gen10 or Gen11 with Windows, VMware or Proxmox on LVM
- You want RAID and HBA drives on the same card (mixed mode)
- One shared Smart Storage Battery for every controller in the box
Dell
Hardware RAID
Dell PERC
H730P · H740P · H755 · H965i · 2 to 8 GB NV cache with battery
Choose it if
- You run PowerEdge 13G to 16G with Windows, Hyper-V or VMware
- You want the biggest cache in the refurbished market: 8 GB on the H740P
- You need HBA mode (H730P) or eHBA (H740P) without a second card
HBA
IT / pass-through
HBA or IT mode
HBA330 · HBA355i · H240ar · MR216i · or a RAID card in HBA mode · no cache
Choose it if
- TrueNAS, Proxmox ZFS, Ceph, vSAN, Storage Spaces Direct or Nutanix will own the disks
- You want pools that import on any other server
- You want SMART, TRIM and checksums to reach the OS
Ask what a server's RAID controller does and most people answer "RAID". The more useful question in 2026 is whether the controller should be doing RAID at all. Windows, Hyper-V, VMware with VMFS and most databases still run best on a hardware controller with a battery-backed write cache. ZFS, Ceph, vSAN, Storage Spaces Direct and Nutanix insist on the opposite: a plain host bus adapter that hands every disk to the software untouched. Put the wrong mode under the wrong software and you get silent write-through performance, hidden disk errors, or a pool that cannot be moved to another machine.
HPE and Dell each ship a full family of controllers that can play either role. HPE calls its RAID cards Smart Array and its pass-through cards Smart HBAs; Dell calls them PERC and HBA. The naming differs, the firmware differs, the batteries differ, and the way each vendor lets you mix RAID and pass-through on one card differs more than the spec sheets suggest.
This guide compares HPE Smart Array, Dell PERC and HBA or IT mode on cache, battery, supported modes, performance, software compatibility, management and price, with a workload-by-workload recommendation. For the model-by-model detail behind each vendor, see our HPE Smart Array guide and Dell PERC guide; for choosing the RAID level itself, see RAID levels explained.
Hardware RAID and HBA mode: what actually changes
A hardware RAID controller presents the operating system with one logical drive and does the work itself: parity calculation, striping, rebuilds and, most importantly, write caching. When a write arrives it is stored in the controller's DRAM cache, protected by a battery or supercapacitor, and acknowledged to the application in a fraction of a millisecond. The physical disks are written later, in full stripes, which hides the RAID 5 and RAID 6 write penalty almost completely.
An HBA does none of that. It passes each disk through unchanged, so the operating system or storage software sees the real drives with their serial numbers, SMART data and TRIM support. Parity, mirroring, checksums and caching are done in software using system RAM and, for synchronous writes, a dedicated log or journal SSD. Writes are acknowledged only when the disk or log device confirms them.

The same write, two paths. Hardware RAID acknowledges from protected cache in about 0.1 ms; HBA mode acknowledges after the disk or a mirrored log device confirms, which is why ZFS and Ceph care so much about a fast log SSD.
Hardware RAID
0.1 to 0.3 ms
Typical cached write latency on a Smart Array or PERC with write-back enabled and a healthy battery
HBA + HDD, no log
5 to 12 ms
Synchronous write latency when every commit must reach a spinning disk
HBA + enterprise SSD log
0.2 to 0.5 ms
ZFS with a power-loss-protected SLOG, or Ceph with an NVMe journal; close to hardware RAID
Smart Array, PERC and HBA mode compared
The three options below are the ones a refurbished ProLiant or PowerEdge buyer will meet. The specifications quoted are for the controllers that ship in Gen9 to Gen11 and 13G to 16G servers; the family map further down lists each generation in detail.

HPE Smart ArrayP440ar · P408i-a · P816i-a · SR932i-p
Hardware RAIDHPE's RAID controllers for ProLiant servers, built on Microchip (formerly Adaptec and PMC) silicon. Gen9 cards such as the P440ar run in either RAID or HBA mode for the whole controller; Gen10 cards such as the P408i-a and P816i-a add mixed mode, so individual drives can be passed through while others sit in a RAID set. Gen11 splits into SR (Microchip, mixed mode, 8 GB cache on the SR932i-p) and MR (Broadcom MegaRAID) lines. All use one shared 96 W Smart Storage Battery or hybrid capacitor to protect the cache.
Cache
2 GB (P440ar, P408i) · 4 GB (P816i) · 8 GB (SR932i-p)
RAID levels
0, 1, 10, 5, 50, 6, 60, 1 ADM, 10 ADM
Modes
RAID, HBA (Gen9) or mixed per drive (Gen10+)
Cache protection
Shared 96 W Smart Storage Battery, up to 20 devices
Management
SSA, ssacli, iLO, Intelligent Provisioning

Advantages
- Mixed mode on Gen10: boot mirror on RAID 1 plus pass-through disks for ZFS or Ceph on one card
- One battery protects every controller in the server and reports health in iLO
- SSD Smart Path accelerates reads on SSD volumes; arrays migrate forward between Smart Array generations
- Mature command line (ssacli) that scripts well and is readable from the OS
- P408i-a and P816i-a are plentiful and cheap on the refurbished market
Limitations
- HBA mode on Gen9 is controller-wide: the whole P440ar is either RAID or HBA
- Firmware and SSA downloads often need an HPE entitlement
- The Smart Storage Battery is a wear item: budget a replacement every 3 to 5 years
- Gen11 MR and SR arrays are not interchangeable
- Embedded B140i and S100i are software RAID, not Smart Array

Dell PERCH730P · H740P · H755 · H965i
Hardware RAIDDell's PowerEdge RAID Controllers, built on Broadcom (LSI) MegaRAID silicon. The 13G-era H730P carries 2 GB of non-volatile cache and can be switched to a full HBA mode; the 14G H740P raises the cache to 8 GB and adds Enhanced HBA (eHBA) from firmware 50.5, which allows RAID 0, 1 and 10 volumes alongside non-RAID disks. The 15G H755 moves to PCIe 4.0 and tri-mode (SAS, SATA and NVMe), and the 16G H965i adds 24G SAS. Each PERC protects its cache with its own battery.
Cache
1 GB (H730) · 2 GB (H730P) · 8 GB (H740P, H755, H965i)
RAID levels
0, 1, 5, 6, 10, 50, 60
Modes
RAID or HBA (H730P) · RAID or eHBA (H740P+)
Cache protection
Per-controller battery backup unit to flash
Management
PERCCLI, OpenManage, iDRAC, Lifecycle Controller


Advantages
- 8 GB cache on the H740P, four times the P408i-a, for the same refurbished money
- Foreign configuration import moves arrays between PERC generations reliably
- MegaRAID tooling is cross-platform: PERCCLI, StorCLI and iDRAC Redfish
- H755 and H755N bring hardware RAID to NVMe on 15G and 16G servers
- Dell firmware is freely downloadable without a support contract
Limitations
- No true pass-through on the H740P: eHBA still runs a RAID stack in front of the disks
- Separate battery per controller, each a wear item with its own alerts
- H330 and H345 have no cache at all and should never host parity RAID
- PERC and Smart Array arrays are not readable by each other
- H730P HBA mode support on hyperconverged HCLs is version-specific; check before buying

HBA or IT modeHBA330 · HBA355i · H240ar · MR216i · or a RAID card in HBA mode
Pass-throughA host bus adapter connects drives to the server and does nothing else. Dell's HBA330 and HBA355i, HPE's H240ar and the Gen11 MR216i are purpose-built pass-through cards with no cache and no RAID logic; "IT mode" is the Broadcom term for the same firmware state on a MegaRAID chip. Many RAID cards can be switched into an equivalent mode. The software above, whether ZFS, Ceph, vSAN, Storage Spaces Direct or Nutanix, then owns every disk, computes its own redundancy, checksums every block and caches in system RAM.
Cache
None on the card; system RAM and a log SSD instead
RAID levels
Whatever the software offers: mirrors, RAID-Z, erasure coding
Modes
Pass-through only; every disk visible by serial
Cache protection
Not needed; power-loss protection on the log SSD instead
Management
smartctl, sas3ircu, storcli, the storage software itself


Advantages
- Required or strongly recommended by TrueNAS, Proxmox ZFS, Ceph, vSAN, S2D and Nutanix
- Pools and OSDs import on any server with any HBA; no controller lock-in
- End-to-end checksums catch silent corruption that hardware RAID cannot see
- Nothing to replace: no battery, no cache module, no licence
- Cheapest option: HBA330 and HBA355i trade for a fraction of a cached PERC
Limitations
- Synchronous writes to HDD are slow without a power-loss-protected log SSD
- Parity and checksums cost CPU and RAM: plan 1 GB of RAM per TB for ZFS
- No boot RAID without the operating system's own mirroring
- Windows and VMFS gain nothing from an HBA and lose the write cache
- Rebuild and scrub behaviour depends on the software, not on a controller you can swap
The modes, and what each vendor really means by them
The words "HBA mode" and "pass-through" are used loosely, and the differences matter when a hyperconverged vendor audits your configuration. There are four distinct states a drive can be in behind a Smart Array or PERC.
| Drive state | What the OS sees | HPE Smart Array | Dell PERC | Safe for ZFS, Ceph, vSAN? |
|---|---|---|---|---|
| RAID logical drive | One virtual disk, cache in front | All models | All models | No: software cannot see the disks |
| Controller-wide HBA mode | Every physical disk, no cache, no RAID | P440ar, P840ar (Gen9), H240ar always | H730, H730P, H330; HBA330 and HBA355i always | Yes: the state vendors certify |
| Mixed mode / eHBA | Some disks as RAID volumes, others pass-through | P408i, P816i, SR Gen11: true per-drive pass-through | H740P, H745, H755, H965i: eHBA (RAID 0/1/10 plus non-RAID disks) | Yes on HPE; usually on Dell, check the HCL |
| Single-disk RAID 0 per drive | One virtual disk per physical disk, cache in front | Possible on any model | Possible on any model | No: hides errors and TRIM, pool not portable |
The RAID 0 trap. Creating a one-disk RAID 0 for every drive looks like pass-through and is the most common mistake we see on refurbished servers running TrueNAS or Proxmox. The controller still caches writes and reorders them, still hides SMART and TRIM, and still stamps the disk with metadata only that controller family can read. When the controller fails, the pool cannot be imported elsewhere. Use real HBA mode, mixed mode or an HBA card instead.
Controller families by server generation
Which controller you get is largely decided by which generation of server you buy. The map below pairs HPE and Dell generations with their RAID and HBA options; our DL380 Gen9 vs Gen10 vs Gen11 and R640 vs R650 vs R660 guides cover the servers themselves.

Cache sizes are flash-backed write cache. Embedded SATA controllers (B140i, S100i, VROC) are driver-based software RAID and are not a substitute for a hardware card.
| Controller | Server generation | Cache | Interface | Pass-through option | Refurbished price |
|---|---|---|---|---|---|
| HPE | |||||
| Smart Array P440ar | Gen9 | 2 GB FBWC | 12G SAS, PCIe 3.0 x8 | HBA mode, controller-wide | About QAR 270 |
| Smart HBA H240ar | Gen9 | None | 12G SAS, PCIe 3.0 x8 | Always pass-through (RAID 0/1/5 optional) | About QAR 150 to 250 |
| Smart Array P408i-a SR Gen10 | Gen10, Gen10 Plus | 2 GB FBWC | 12G SAS, 8 lanes, PCIe 3.0 x8 | Mixed mode per drive | QAR 520 |
| Smart Array P816i-a SR Gen10 | Gen10, Gen10 Plus | 4 GB FBWC | 12G SAS, 16 lanes, PCIe 3.0 x8 | Mixed mode per drive | About QAR 800 to 1,100 |
| Smart Array E208i-a | Gen10 | None | 12G SAS, 8 lanes | Mixed mode; RAID 0/1/5/10 only | About QAR 200 to 300 |
| HPE SR932i-p Gen11 | Gen11 | 8 GB FBWC | 24G SAS, NVMe tri-mode, PCIe 4.0 x16 | Mixed mode per drive | New only, quote |
| HPE MR216i-o / -p Gen11 | Gen11 | None | 24G SAS, NVMe tri-mode | Pass-through class (MegaRAID) | New only, quote |
| Dell | |||||
| PERC H330 | 13G, 14G | None | 12G SAS, PCIe 3.0 x8 | Non-RAID disks, HBA mode | QAR 180 to 420 |
| PERC H730 / H730P | 13G | 1 GB / 2 GB NV | 12G SAS, PCIe 3.0 x8 | HBA mode, controller-wide | QAR 240 (H730) |
| HBA330 | 13G, 14G | None | 12G SAS, PCIe 3.0 x8 | IT mode, always | About QAR 250 to 400 |
| PERC H740P | 14G | 8 GB NV | 12G SAS, PCIe 3.1 x8 | eHBA from firmware 50.5 | QAR 570 |
| HBA355i | 14G, 15G, 16G | None | 12G SAS, PCIe 4.0 x8 | IT mode, always | About QAR 350 to 500 |
| PERC H755 / H755N | 15G, 16G | 8 GB NV | 12G SAS and NVMe tri-mode, PCIe 4.0 x8 | eHBA | About QAR 1,500 to 2,500 |
| PERC H965i | 16G | 8 GB NV | 24G SAS and NVMe, PCIe 4.0 | eHBA | New only, quote |
Cache, batteries and the write-through trap
The cache is the whole reason to buy a hardware RAID controller, and the battery is what makes the cache safe to use. When the battery fails, is missing or is still charging after a cold start, both vendors quietly switch the cache to write-through. The array keeps working, but every write now waits for the disks, and a RAID 5 or RAID 6 set on hard drives can lose 80 to 90 percent of its random write performance overnight. On a refurbished server this is the single most common cause of "the new server is slower than the old one".
| Cache protection | HPE Smart Array | Dell PERC | HBA mode |
|---|---|---|---|
| Component | 96 W Smart Storage Battery (Li-ion) or Smart Storage Hybrid Capacitor, cabled to the system board | Battery backup unit (Li-ion) plugged into each controller | None; power-loss protection lives on the log or journal SSD |
| Controllers protected | Up to 20 devices from one battery | One per controller | Not applicable |
| What happens on power loss | Battery powers the controller long enough to flush DRAM to on-card flash | Battery powers the controller long enough to flush DRAM to on-card flash | In-flight data in RAM is lost; the log SSD preserves committed sync writes |
| Typical service life | 3 to 5 years; health shown in iLO and SSA | 3 to 5 years; health shown in iDRAC and OpenManage | Log SSD endurance: check DWPD, typically 3 to 10 DWPD for a SLOG |
| Behaviour when degraded | Write cache disabled, logical drives drop to write-through, iLO warning | Write cache disabled, virtual disks drop to write-through, iDRAC warning | No change; sync-write latency depends on the log device |
| Replacement cost | QAR 240 to 290 | About QAR 150 to 300 depending on model | Enterprise SSD with PLP: QAR 400 to 900 |
Check the cache state before judging performance. On HPE, run ssacli ctrl all show status or look at Storage in iLO; on Dell, perccli /c0 show or the Storage page in iDRAC. "Cache Status: Temporarily Disabled" or "Write Policy: Write Through" means the battery is the problem, not the disks. For ZFS, the equivalent check is whether a SLOG is present and whether the dataset is set to sync=standard. Our refurbished server inspection checklist includes both.
Performance: where each approach wins
On the same twelve-drive chassis, the three approaches land in very different places depending on the workload. The bars below are indicative figures for a Gen10 or 14G server with twelve 10K SAS drives and, for the HBA case, ZFS with an enterprise SSD log device.
Random 4K write latency, 12 × 10K SAS, lower is better
Sequential read, 12 × 10K SAS, RAID 6 or RAID-Z2, higher is better
Indicative figures from vendor documentation and our own bench tests on refurbished Gen10 and 14G systems. Sequential throughput is limited by the drives; random write latency is decided almost entirely by where the write is acknowledged.
Three patterns hold across hardware. First, cached hardware RAID and software RAID with a good log device are close on latency and identical on sequential throughput; the gap opens only when the cache is off or the log device is missing. Second, the CPU cost of software RAID is real but small on modern servers: ZFS RAID-Z2 on twelve drives uses roughly one core's worth of a Gold 6140 at full write speed, and Ceph's erasure coding somewhat more. Third, on all-NVMe systems the picture flips: a tri-mode controller's x8 or x16 host link and RAID engine become the ceiling at around one to two million IOPS, while the same drives on CPU lanes with VROC or software RAID scale further. For NVMe-heavy builds, see NVMe vs SAS vs SATA SSDs.
Software compatibility: which mode each platform wants
This is the table to settle an argument with. Each platform has a documented preference, and in several cases the vendor will refuse support if you ignore it.
| Platform | Recommended controller mode | HPE option | Dell option | Notes |
|---|---|---|---|---|
| Windows Server, Hyper-V | Hardware RAID, write-back | P408i-a, P816i-a | H730P, H740P, H755 | RAID 10 for Hyper-V and SQL, RAID 6 for file services |
| VMware ESXi with VMFS | Hardware RAID, write-back | P408i-a, P816i-a | H730P, H740P, H755 | Local VMFS datastores depend on the cache for write performance |
| VMware vSAN | Pass-through on HCL | H240ar, P408i mixed mode (HCL listed) | HBA330, HBA355i, H730P HBA mode | Controller and firmware must match the vSAN HCL exactly |
| Proxmox VE on LVM or ext4 | Hardware RAID, write-back | Any Smart Array | Any cached PERC | Simplest path for VMs on local storage |
| Proxmox VE on ZFS | HBA or mixed mode | P408i mixed mode, H240ar | HBA330, HBA355i, H730P HBA mode | Add an enterprise SSD mirror as SLOG for VM workloads |
| TrueNAS CORE and SCALE | HBA only | H240ar, P440ar HBA mode | HBA330, HBA355i | iXsystems explicitly warns against RAID cards and RAID 0 per disk |
| Ceph, including Proxmox Ceph | HBA only | H240ar, P408i mixed mode | HBA330, HBA355i | One OSD per disk; NVMe DB/WAL device recommended for HDD OSDs |
| Storage Spaces Direct | HBA only, SES-capable | H240ar, Gen10 HBA parts | HBA330, HBA355i | Microsoft rejects RAID controllers outright for S2D |
| Nutanix AHV | Pass-through | Certified HBA parts only | HBA330, HBA355i | Factory-certified nodes only in practice |
| Veeam or Proxmox Backup repository | Hardware RAID 6 or ZFS RAID-Z2 | P816i-a with 12 LFF | H740P with 12 LFF | Either works; hardware RAID is simpler, ZFS adds checksums |
| Bare-metal SQL Server, Oracle, PostgreSQL | Hardware RAID 10, write-back | P816i-a | H740P, H755 | Cache absorbs log writes; separate logs from data |
For the broader comparison of hardware RAID with software-defined storage, see Ceph vs ZFS vs hardware RAID, and for a ZFS build see our TrueNAS business storage guide.

Pick the mode from the software first. Most refurbished Gen10 and 14G servers can do both with the controller they already have.
Management, monitoring and firmware
| Task | HPE Smart Array | Dell PERC | HBA mode |
|---|---|---|---|
| Configure arrays | Smart Storage Administrator (SSA) in Intelligent Provisioning or from the OS; ssacli | PERC BIOS (F2), Lifecycle Controller, OpenManage Server Administrator; perccli | The storage software: TrueNAS UI, Proxmox, Ceph tools |
| Out-of-band health | iLO 4/5/6 Storage page, battery and cache status included | iDRAC 8/9 Storage page, battery and cache status included | Drive health only via iLO or iDRAC; pool health from the OS |
| SMART and TRIM | Hidden behind logical drives; visible for mixed-mode HBA drives | Hidden behind virtual disks; visible for non-RAID disks | Fully visible to the OS |
| Moving an array to another server | Arrays migrate to the same or newer Smart Array generation; not to MR Gen11 | Foreign configuration import on the same or newer PERC generation | Pool imports on any server with any HBA |
| Firmware access | Many SPP and controller downloads need an active HPE entitlement | Freely downloadable from Dell support | Dell and Broadcom firmware freely available; HPE H240 needs entitlement |
| Alerting | iLO email and SNMP, Insight Remote Support | iDRAC email and SNMP, OpenManage Enterprise | ZED for ZFS, Ceph manager alerts, S2D health service |
The practical difference for a refurbished buyer is firmware. Dell publishes PERC and HBA firmware openly, so a 14G server can be brought current by anyone. HPE gates many Smart Array firmware packages behind a warranty or contract, so a Gen9 or Gen10 should arrive from the seller with the controller already on a current release and the Smart Storage Battery tested. Our iDRAC vs iLO guide covers the management side in more depth.
Price and value in 2026
Controllers are cheap relative to the server, which is why running the wrong mode to save QAR 300 is a false economy. Current ServerDove prices for the parts in stock, with indicative figures for the rest.
| Part | Role | Price | Buy it when |
|---|---|---|---|
| HPE Smart Array P408i-a SR Gen10, 2 GB | Hardware RAID, mixed mode | QAR 520 | Any Gen10 host for Windows, VMware or Proxmox; also ZFS via mixed mode |
| HPE Smart Storage Battery 96 W (MC96) | Cache protection | QAR 240 to 290 | Whenever iLO shows the battery degraded or missing |
| HPE Smart Array P440ar, 2 GB | Hardware RAID or HBA mode | About QAR 270 | Gen9 hosts; switch to HBA mode for TrueNAS or Ceph |
| Dell PERC H740P, 8 GB | Hardware RAID, eHBA | QAR 570 | 14G hosts running databases or heavy VM write loads |
| Dell PERC H730, 1 GB | Hardware RAID or HBA mode | QAR 240 | 13G hosts; cheapest cached PERC with a true HBA mode |
| Dell PERC H330 | No cache, non-RAID capable | QAR 180 to 420 | Boot mirrors and HBA duty only; never parity RAID |
| Dell HBA330 / HBA355i | IT mode HBA | About QAR 250 to 500 | vSAN, S2D, Ceph and TrueNAS on 13G to 16G |
| Enterprise SSD with power-loss protection, 400 to 960 GB | ZFS SLOG or Ceph DB/WAL | QAR 400 to 900 | Any HBA build with synchronous writes: VMs, NFS, databases |
Put differently: a Gen10 server with a P408i-a already in it costs nothing extra to run in mixed mode, and a 14G server with an H740P needs only an HBA330 (about QAR 300) to become a fully supported vSAN or TrueNAS node while keeping the H740P for a boot mirror. The expensive mistake is buying a cache-less H330 or E208i and then running RAID 5 on it.
HPE controllers and batteries in stock
HPE servers and parts →Dell PERC controllers in stock
Dell servers and parts →The controller is not the decision. The software is. Choose what will manage the disks, then put the controller into the mode that software documents, and the vendor arguments disappear.
Which setup for which workload
Virtualisation host, 10 to 60 VMs on local disks
Hyper-V, ESXi with VMFS or Proxmox on LVM: cached hardware RAID 10 on SAS SSD or 10K drives. Proxmox on ZFS: HBA or mixed mode with a mirrored SLOG.
SQL Server, ERP and transactional databases
Write-back cache absorbs log flushes. P816i-a or H740P in RAID 10, battery verified, separate volumes for data, logs and tempdb.
TrueNAS or ZFS file server and NAS
HBA only. HBA330, HBA355i or H240ar, or a P440ar / H730P switched to HBA mode. Mirrors for VM storage, RAID-Z2 for archives, SLOG for NFS and iSCSI.
Proxmox or Ceph cluster, three or more nodes
One OSD per pass-through disk, NVMe for DB/WAL. Mixed-mode P408i-a or HBA330; keep a small RAID 1 boot volume if the card allows it.
vSAN, Storage Spaces Direct or Nutanix
Pass-through controller from the platform's compatibility list, correct firmware, no cache in the path. HBA330 and HBA355i are the safe Dell choices; HPE nodes need HCL-listed parts.
Backup repository: Veeam, Proxmox Backup Server
Twelve LFF drives in hardware RAID 6 with cache, or ZFS RAID-Z2 on an HBA if you want checksums and snapshots. Both work; pick the one your team can operate.
All-NVMe host: analytics, AI data, scratch
Drives on CPU lanes with VROC, mdadm or ZFS. A tri-mode H755N or SR932i-p only when hardware RAID on NVMe is a policy requirement, and accept the IOPS ceiling.
Small business all-in-one, one server, no storage specialist
Cached hardware RAID 1 or RAID 10 with iLO or iDRAC alerting. It is the configuration least likely to surprise anyone, and the one the vendor's phone support understands.
Switching modes and moving arrays
Changing a controller's mode destroys the metadata on its drives, so it is a build-time decision, not a tuning step. These are the rules for moving between the three worlds.
Smart Array to a same-or-newer Smart Array (P440ar to P408i-a). PERC to a same-or-newer PERC via foreign import (H730P to H740P). Any ZFS pool or Ceph OSD to any server with any HBA. A RAID card from RAID mode to HBA mode after the drives are wiped.
HBA mode back to RAID mode: all drives are reinitialised. H740P eHBA still runs a RAID stack, so vSAN and S2D may require a separate HBA330 instead. HPE SR to MR Gen11 arrays do not migrate. Mixed-mode pass-through drives keep no controller metadata, so they move freely; the RAID volumes on the same card do not.
Smart Array arrays on a PERC, or PERC arrays on a Smart Array. Single-disk RAID 0 volumes imported as plain disks on another controller. Cached RAID volumes read by an HBA. Using a cache-less H330, H345 or E208i for RAID 5 or 6 and expecting production write speed.
Where controllers are heading
Now
Tri-mode
SAS + SATA + NVMe
H755, H965i and SR932i-p put NVMe behind hardware RAID for teams that need it
Growing
HBA by default
ZFS, Ceph, vSAN
Hyperconverged and software-defined platforms now ship most new nodes with pass-through controllers
Arriving
24G SAS, PCIe 5.0
Gen11 · 16G
Double the lane bandwidth; cache sizes settle at 8 GB with hybrid capacitors replacing batteries
Steady
Refurbished market
P408i-a · H740P
Both stay the value picks through 2028 as Gen10 and 14G dominate the second-hand supply
Two trends matter for anyone buying refurbished today. Hardware RAID is not going away, but it is becoming a feature of the Windows and VMware world rather than a universal default, and new storage platforms assume an HBA. At the same time, controller cache has stopped growing at 8 GB while the drives behind it have moved to flash, so the write cache matters less on an all-SSD array than it did on spinning disks. On a Gen10 or 14G server the pragmatic answer is to keep the cached controller, learn its mixed or HBA mode, and let the workload decide drive by drive.
Conclusion
HPE Smart Array and Dell PERC are mature, fast and well supported, and for Windows, Hyper-V, VMware VMFS, Proxmox on LVM and databases a cached controller with a healthy battery remains the right choice. Dell's H740P offers the most cache per riyal; HPE's P408i-a offers the most flexible mixed mode. Neither should be used to fake pass-through with single-disk RAID 0.
For ZFS, Ceph, vSAN, Storage Spaces Direct and Nutanix, the controller should disappear: an HBA330, HBA355i or H240ar, or a Smart Array or PERC in a real HBA or mixed mode, with an enterprise SSD taking over the job the cache used to do. Decide which world the server belongs to before you buy, verify the cache and battery state on arrival, and the storage controller becomes the most boring and reliable part of the machine.
Tell us the software and we will configure the controller to match.
ServerDove supplies tested HPE Smart Array and Dell PERC controllers, HBAs, batteries and enterprise drives in Qatar, and ships refurbished servers with the controller already set to RAID, HBA or mixed mode for your platform.
Common questions
What is the difference between HPE Smart Array and Dell PERC?
Both are hardware RAID controllers with battery-protected write cache. Smart Array is HPE's family for ProLiant servers, built on Microchip silicon and managed with SSA and iLO; PERC is Dell's family for PowerEdge, built on Broadcom MegaRAID silicon and managed with PERCCLI, OpenManage and iDRAC. HPE protects every controller from one shared 96 W Smart Storage Battery; each PERC has its own battery. Arrays created on one family cannot be read by the other.
What is HBA mode, and is it the same as IT mode?
HBA mode means the controller passes each physical disk through to the operating system without RAID or caching. IT mode (Initiator Target) is the Broadcom firmware term for the same state on MegaRAID-based cards such as the Dell HBA330. HPE calls its equivalent HBA mode on Gen9 Smart Arrays and mixed mode on Gen10 and later, where individual drives can be passed through alongside RAID volumes.
Can I use a RAID controller for ZFS or TrueNAS?
Only in a true HBA or mixed mode that presents raw disks. A Smart Array P440ar or PERC H730P in HBA mode, or a P408i-a in mixed mode, is acceptable; a dedicated HBA330, HBA355i or H240ar is better. Never create a single-disk RAID 0 per drive: the controller hides SMART and TRIM, caches writes outside ZFS's control, and the pool cannot be imported on another controller.
Does the Dell PERC H740P support HBA mode?
The H740P does not have a full HBA mode. From firmware 50.5 it supports Enhanced HBA (eHBA) mode, which allows RAID 0, 1 and 10 virtual disks alongside non-RAID disks presented to the host. For platforms that require a certified pass-through controller, such as vSAN or Storage Spaces Direct, Dell's answer is a separate HBA330 or HBA355i.
Why is my RAID array suddenly slow?
The most common cause is the cache falling back to write-through because the battery is failed, missing or still charging. Check the controller status in iLO or iDRAC, or run ssacli or perccli from the OS. A replacement HPE Smart Storage Battery costs QAR 240 to 290; PERC batteries are similar. On ZFS, the equivalent problem is a missing or worn-out SLOG device.
Is hardware RAID still worth it in 2026?
Yes, for Windows, Hyper-V, VMware VMFS, Proxmox on LVM and most databases, where a cached controller gives sub-millisecond write latency with no CPU cost and is the configuration vendors support. It is not worth it, and is often unsupported, under ZFS, Ceph, vSAN, Storage Spaces Direct and Nutanix, which need direct access to the disks.
Which controller should I buy with a refurbished Gen10 or 14G server?
For HPE Gen10, the Smart Array P408i-a (QAR 520) covers hardware RAID and mixed-mode pass-through on one card. For Dell 14G, the PERC H740P (QAR 570) gives 8 GB of cache for RAID workloads, and an HBA330 (about QAR 300) is the right addition if the server will run vSAN, TrueNAS or Ceph. In both cases confirm the battery and firmware state before the server goes into production.





