The Dell H730 Mini’s HBA Mode is NOT What You Think!

If you’re building a ZFS system or performance-sensitive storage server, you’ve probably come across the Dell PERC H730 Mini — a RAID controller found in many Dell R630, R730, and other 13th-gen PowerEdge servers. It’s great since it fits in the integrated slot and doesn’t occupy one of the PCIe expansion slots.

Dell refer to its “HBA mode” as a way to use individual drives without RAID; to passthrough the drives to the OS for use with Ceph or ZFS.

Sounds perfect, right? Not so fast. Literally!

I’ve been trying to figure out what’s actually going on with these cards after this problem reared it’s head with some SAS SSDs (which can push some serious performance numbers).

I’m pretty sure Dell is telling a white lie with their HBA mode. It normally wouldn’t be a problem – but high speed SSDs show it up.

What Dell Calls “HBA Mode” Isn’t the Real Deal

The PERC H730 Mini does not support true HBA/IT mode. The LSI SAS3108 chipset is physically INCAPABLE of IT mode – meaning whichever way you look at it, it’s bad. Instead, when you put the controller into “HBA” or “non-RAID” mode, 1 of 2 things are likely happening:

It wraps each physical disk into a single-disk RAID 0 virtual drive and presents this virtual disk to the OS.
It emulates the drive. This might sound ok, but it still means that every bit is going through the RAID firmware and stack, increasing latency and adding a throttle point. And it doesn’t even emulate everything (like SMART or TRIM).

True IT Mode (HBA):

OS talks directly to the physical disk
Full support for SMART, TRIM, native block alignment
No caching, translation, or abstraction
Ideal for filesystems like ZFS

H730 “HBA” Mode:

OS talks to a virtual disk (single-disk RAID 0)
Controller firmware still intercepts all commands
Limited or partial access to SMART and advanced SSD features
Often hides failures or introduces additional latency

Performance Problems in Practice

Using the H730 Mini in “HBA” mode with SAS SSDs that are capable of >100K IOPS and 1GB/s each? You may find:

You’re not getting the throughput you should on a single drive.
ZFS performance is much worse than it COULD be.
IOPS bottlenecked by firmware queue depth limits – I noticed this particularly when loading the pool up with multiple VMs. Everything just got a bit sluggish.
TRIM, SMART, and drive health data missing

ZFS in particular requires direct drive access to manage redundancy, detect errors, and replace failed disks. If the controller masks this, it undermines ZFS’s entire design.

I’m not 100% sure, but this could also cause problems with portability. Because the H730 is in the middle and potentially doing a RAID0 on each disk, it could cause them to not show up properly in the event that you exported your ZFS pool to another system with a true HBA, for example. This is because of RAID metadata and other strange partitions and alignments that hardware raid makes.