Last month I lost a drive in my TrueNAS mirror. WD Red, three years old, SMART warnings I’d been ignoring for two weeks. The rebuild took 14 hours on spinning rust, and the whole time I was thinking: if the second drive goes, that’s 8TB of media and backups gone.
That rebuild forced me to actually research what I was putting in my NAS instead of just grabbing whatever was on sale. Turns out, picking the right drives for ZFS matters more than most people realize — and the wrong choice can cost you data or performance.
Here’s what I learned, what I’m running now, and what I’d buy if I were building from scratch today.
CMR vs. SMR: This Actually Matters for ZFS
Before anything else — check if your drive uses CMR (Conventional Magnetic Recording) or SMR (Shingled Magnetic Recording). ZFS and SMR don’t get along. SMR drives use overlapping write tracks to squeeze in more capacity, which means random writes are painfully slow. During a resilver (ZFS’s version of a rebuild), an SMR drive can take 3-4x longer than CMR.
WD got caught shipping SMR drives labeled as NAS drives back in 2020 (the WD Red debacle). They’ve since split the line: WD Red Plus = CMR, plain WD Red = SMR. Don’t buy the plain WD Red for a NAS. I made this mistake once. Never again.
Seagate’s IronWolf line is all CMR. Toshiba N300 — also CMR. If you’re looking at used enterprise drives (which I’ll get to), they’re all CMR.
The Drives I’d Actually Buy Today
For Bulk Storage: WD Red Plus 8TB
The WD Red Plus 8TB (WD80EFPX) is what I’m running right now. 5640 RPM, CMR, 256MB cache. It’s not the fastest drive, but it runs cool and quiet — important when your NAS sits in a closet six feet from your bedroom.
Price per TB on the 8TB sits around $15-17 at time of writing. The sweet spot for capacity vs. cost. Going to 12TB or 16TB drops the per-TB price slightly, but the failure risk per drive goes up — a single 16TB drive failing is a lot more data at risk during rebuild than an 8TB.
I run these in a mirror (RAID1 equivalent in ZFS). Two drives, same data on both. Simple, reliable, and rebuild time is reasonable. Full disclosure: affiliate link.
If You Prefer Seagate: IronWolf 8TB
The Seagate IronWolf 8TB (ST8000VN004) is the other solid choice. 7200 RPM, CMR, 256MB cache. Faster spindle speed means slightly better sequential performance, but also more heat and noise.
Seagate includes their IronWolf Health Management software, which hooks into most NAS operating systems including TrueNAS. It gives you better drive health telemetry than standard SMART. Whether that’s worth the slightly higher price depends on how paranoid you are about early failure detection. (I’m very paranoid, but I still went WD — old habits.)
Both drives have a 3-year warranty. The IronWolf Pro bumps that to 5 years and adds rotational vibration sensors (matters more in 8+ bay enclosures). For a 4-bay homelab NAS, the standard IronWolf is enough. Full disclosure: affiliate link.
Budget Option: Used Enterprise Drives
Here’s my hot take: refurbished enterprise drives are underrated for homelabs. An HGST Ultrastar HC320 8TB can be found for $60-80 on eBay — roughly half the price of new consumer NAS drives. These were built for 24/7 operation in data centers. They’re louder (full 7200 RPM, no acoustic management), but they’re tanks.
The catch: no warranty, unknown hours, and you’re gambling on remaining lifespan. I run one pool with used enterprise drives and another with new WD Reds. The enterprise drives have been fine for two years. But I also keep backups, because I’m not an idiot.
SSDs in TrueNAS: SLOG, L2ARC, and When They’re Worth It
ZFS has two SSD acceleration features that confuse a lot of people: SLOG (write cache) and L2ARC (read cache). Let me save you some research time.
SLOG (Separate Log Device)
SLOG moves the ZFS Intent Log to a dedicated SSD. This only helps if you’re doing a lot of synchronous writes — think iSCSI targets, NFS with sync enabled, or databases. If you’re mostly streaming media and storing backups, SLOG does nothing for you.
If you DO need a SLOG, the drive needs high write endurance and a power-loss protection capacitor. The Intel Optane P1600X 118GB is the gold standard here — extremely low latency and designed for exactly this workload. They’re getting harder to find since Intel killed the Optane line, but they pop up on Amazon periodically. Full disclosure: affiliate link.
Don’t use a consumer NVMe SSD as a SLOG. If it loses power mid-write without a capacitor, you can lose the entire transaction log. That’s your data.
L2ARC (Level 2 Adaptive Replacement Cache)
L2ARC is a read cache on SSD that extends your ARC (which lives in RAM). The thing most guides don’t tell you: L2ARC uses about 50-70 bytes of RAM per cached block to maintain its index. So adding a 1TB L2ARC SSD might eat 5-10GB of RAM just for the metadata.
Rule of thumb: if you have less than 64GB of RAM, L2ARC probably hurts more than it helps. Your RAM IS your cache in ZFS — spend money on more RAM before adding an L2ARC SSD. I learned this the hard way on a 32GB system where L2ARC actually slowed things down.
If you do have the RAM headroom and want L2ARC, any decent NVMe drive works. I’d grab a Samsung 990 EVO 1TB — good endurance, solid random read performance, and the price has come down a lot. Full disclosure: affiliate link.
What My Actual Setup Looks Like
For context, I run TrueNAS Scale on an older Xeon workstation with 64GB ECC RAM. Here’s the drive layout:
Pool: tank (main storage)
Mirror: 2x WD Red Plus 8TB (WD80EFPX)
Pool: fast (VMs and containers)
Mirror: 2x Samsung 870 EVO 1TB SATA SSD
No SLOG (my workloads are async)
No L2ARC (64GB RAM handles my working set)Total usable: ~8TB spinning + ~1TB SSD. The SSD pool runs my Docker containers and any VM images. Everything else — media, backups, time machine targets — lives on the spinning rust pool.
This separation matters. ZFS performs best when pools have consistent drive types. Mixing SSDs and HDDs in the same vdev is asking for trouble (the pool performs at the speed of the slowest drive).
ECC RAM: Not Optional, Fight Me
While we’re talking about TrueNAS hardware — get ECC RAM. Yes, TrueNAS will run without it. No, that doesn’t mean you should.
ZFS checksums every block, which means it can detect corruption. But if your RAM flips a bit (which non-ECC RAM does more often than you’d think), ZFS might write that corrupted data to disk AND update the checksum to match. Now you have silent data corruption that ZFS thinks is fine. With ECC, the memory controller catches and corrects single-bit errors before they hit disk.
Used DDR4 ECC UDIMMs are cheap. A 32GB kit runs $40-60 on eBay. There’s no excuse not to use it if your board supports it. If you’re building a new system, look at Xeon E-series or AMD platforms that support ECC.
How to Check What You Already Have
Already running TrueNAS? Here’s how to check your drive health before something fails:
# List all drives with model and serial
smartctl --scan | while read dev rest; do
echo "=== $dev ==="
smartctl -i "$dev" | grep -E "Model|Serial|Capacity"
smartctl -A "$dev" | grep -E "Reallocated|Current_Pending|Power_On"
done
# Quick ZFS pool health check
zpool status -vWatch for Reallocated_Sector_Ct above zero and Current_Pending_Sector above zero. Those are your early warning signs. If both are climbing, start shopping for a replacement drive now — don’t wait for the failure like I did.
The Short Version
If you’re building a TrueNAS box in 2026:
- Bulk storage: WD Red Plus or Seagate IronWolf. CMR only. 8TB is the sweet spot for price per TB.
- SLOG: Only if you need sync writes. Intel Optane if you can find one. Otherwise skip it.
- L2ARC: Only if you have 64GB+ RAM to spare. Any NVMe SSD works.
- RAM: ECC or go home. At least 1GB per TB of storage, 32GB minimum.
- Budget move: Used enterprise HDDs + new SSDs for VM pool. Loud but reliable.
Don’t overthink it. Get CMR drives, get ECC RAM, keep backups. Everything else is optimization.
If you found this useful, check out my guides on self-hosting Ollama on your homelab, backup and recovery for homelabs, and setting up Wazuh and Suricata for home security monitoring.
Join https://t.me/alphasignal822 for free market intelligence.
📧 Get weekly insights on security, trading, and tech. No spam, unsubscribe anytime.
Leave a Reply