The hard part of buying a first NAS is not choosing a model. It is working out which decisions you can revise later and which ones you are stuck with for the life of the machine. Memory, network cards and even drives can usually be changed. Bay count and the physical enclosure cannot. Get that ordering right and a modest first purchase lasts years; get it wrong and the second purchase arrives within eighteen months.
Start from the job, not the model
Three jobs dominate first NAS purchases, and they pull hardware in different directions.
A backup target is write-mostly, mostly sequential, and idle for twenty-three hours a day. It needs capacity and reliability, not speed. This is the least demanding job and the one most over-specified.
A file and media server is read-mostly and sequential. It cares about sustained throughput and network headroom, and it is where the gap between a gigabit port and anything faster becomes obvious.
An application host running containers, a photo index, or a database is random-IO heavy and memory-hungry. This is the job that punishes the fixed, non-upgradable RAM in cheap appliances, and the one most likely to justify building rather than buying.
Write down which of the three you actually need before comparing any two products. Most of the disagreement in NAS buying advice is people answering different questions.
Browse all NAS hardware guides to plan the purchase, or explore NAS hardware topics for related reading on drives, networking and array choices.
Bay count is the decision you cannot undo
Adding capacity to a running array means either replacing every drive in a group with a larger one, or adding a whole new group. Both are expensive and slow. Buying an enclosure with more bays than today’s plan requires is by far the cheapest form of future-proofing available.
| Bays | Realistic job | Redundancy available | Main limitation |
|---|---|---|---|
| 1 | Single-disk backup target, second copy of something else | None | A drive failure is total loss; only safe as one copy of several |
| 2 | Documents, photos, small household file server | Mirror | Half the raw capacity is redundancy, and it stays half forever |
| 4 | General household server, first sensible parity array | Single parity or two mirror pairs | Single parity across four large drives means a long, exposed rebuild |
| 6 to 8 | Media library, small office, mixed workloads | Double parity | Rebuild windows grow with drive size, not with bay count |
| 12+ | Multi-group arrays, capacity growth in stages | Double parity per group | Power, cooling, controller lanes and noise all become real design problems |
Two figures matter more than the headline bay count. The first is how much usable space remains after redundancy: a two-bay mirror gives you one drive of usable capacity no matter which drives you fit. The second is the rebuild window, which scales with the size of each drive rather than with the number of them. A four-bay box full of 20 TB drives has a far longer exposed rebuild than a four-bay box of 4 TB drives, which is the real argument for double parity as capacities climb.
Buy an appliance or build one
An appliance NAS from Synology, QNAP, Asustor or similar arrives with a supported operating system, low idle power, a small footprint and a warranty on the whole machine. What you give up is choice: the CPU, the network port and usually the memory ceiling are fixed at purchase.
Building your own trades that convenience for control over PCIe slots, ECC memory support, networking and the chassis. Identify the expansion an appliance cannot provide before ordering parts. A larger processor alone does not solve a shortage of bays or controller connections.
The NAS Drive Bay and PSU Calculator turns a capacity target after parity into a bay count and an estimated startup power budget, with an example PCIe allocation. It excludes filesystem overhead and free-space reserves; use it to compare a shortlist, then check the component manuals.
Chassis bays, backplanes and room to expand
Count occupied bays and future bays separately. A chassis that fits today’s array exactly leaves no empty slot for another disk. Expansion support depends on the storage platform and layout, so confirm that plan before assuming unused bays automatically become usable capacity.
Check whether the backplane uses direct connections or a SAS expander, which connectors it needs, and whether hot swapping is supported by the complete system. The chassis, backplane, cables and controller must agree. TrueNAS’s hardware guide covers SAS expanders and disk connectivity; use the enclosure manual for its wiring. Reserve airflow through the drive stack as well as room around the motherboard.
HBAs and PCIe lanes: check electrical width
When onboard SATA ports are insufficient, a host bus adapter can expose individual drives to the operating system. TrueNAS recommends suitable HBAs for direct disk access instead of hardware RAID volumes. Check the supported firmware mode and driver for the exact card; do not assume every RAID controller can become an HBA.
A physically long PCIe slot may have fewer electrical lanes than its length suggests. List the HBA, network card and NVMe devices together, then consult the motherboard’s slot-sharing table. Check which slots connect through the chipset and whether populating an M.2 socket disables a SATA port or reduces another slot’s width. These are board-specific constraints.
Broadcom’s HBA specifications identify host interfaces and drive connections. Match both to the board and backplane. Keep the NIC in the same budget: a card that fits physically still needs a supported slot and enough airflow. The NAS hardware transfer diagnosis explains how an undersized connection can constrain several drives together.
Drives: count first, capacity second
Decide how many drives the array will hold before deciding how large each one is. The redundancy math depends on the count; the capacity is then simple arithmetic against your target.
Two drive specifications matter more than the marketing tier. The first is recording technology. TrueNAS’s hardware guide advises avoiding SMR drives because they can perform poorly during writes and resilvers. Western Digital’s cited Red Plus brief lists CMR for every model in its table. Read CMR or SMR for NAS arrays for the buying decision, then use the CMR/SMR model lookup to check a documented model number.
The second is the annualised workload rating, which is the amount of data the manufacturer expects to be read from or written to the drive per year. Western Digital rates WD Red Plus at 180 TB per year with a stated MTBF of one million hours, a non-recoverable read error rate below 1 in 10^14 bits, and a three-year limited warranty. Desktop drives are routinely rated at a fraction of that. For a machine that is powered on continuously, the workload figure is the specification that separates a NAS drive from a cheaper one with the same capacity on the label.
Reasonable starting points are the WD Red Plus range and the Seagate IronWolf range. Both ship in many capacities, so buy the capacity your bay count and target require rather than the one at the top of a search results page.
Memory, and why the cheap boxes hit a wall
Appliance memory is often soldered. A Synology DS124 ships with 1 GB of non-ECC DDR4; a DS223 ships with 2 GB. That is ample for file sharing and backup, and it is the constraint that bites the moment you add a photo index, several containers, or a database.
ECC memory is a separate question from capacity. It can correct single-bit memory errors; it does not replace backups or protect against every source of corruption. TrueNAS documents its benefits for storage systems. Verify ECC operation across the CPU, motherboard and modules against their support lists, and distinguish supported ECC operation from a board merely accepting an ECC DIMM. Check memory capacity and spare slots against the planned software as a separate decision.
The network port is the ceiling nobody reads
A single gigabit port carries 1000 Mb/s, which is 125 MB/s before any protocol overhead and roughly 113 MB/s of file data in practice. That number is the ceiling on every transfer to and from the machine, regardless of how many drives sit behind it. A four-drive array will saturate a gigabit link without effort.
Two consequences follow. First, if large-file speed matters to you, check the port before checking the CPU. Second, link aggregation does not fix it. Synology’s own knowledge base article on exactly this question states that total network bandwidth “will only increase if there are multiple clients”, because each client is answered by one of the bonded interfaces. Bonding two gigabit ports buys you two clients at a gigabit each, not one client at two. If transfers are already slower than that ceiling, work through NAS slow transfer speeds before buying anything.
Power, spin-up and the UPS
Steady-state draw is not the number that sizes a power supply. Western Digital’s brief lists 12 V peak current of 1.9 A for the 12 TB WD120EFGX against average read/write consumption of 8.8 W and idle consumption of 6.1 W. That peak works out to roughly 23 W on the 12 V rail per drive during spin-up, against under 9 W once running. Eight drives starting simultaneously is a very different load from eight drives working.
Appliances hide this: Synology quotes 17.3 W during access for the DS223 with drives fitted, and 4.1 W once those drives hibernate. Builds do not hide it, and staggered spin-up exists specifically to spread that peak over several seconds.
Use the PSU’s 12 V rail rating and connector limits as well as its total wattage. Allow for future drives, fans and expansion cards. The same WD brief lists 1.84 A peak and 6.3 W read/write for the other 12 TB model, WD120EFBX: capacity alone is not a power specification. Count simultaneous startup unless the controller and drives explicitly support the staggered startup arrangement you intend to use. Follow the enclosure’s cooling requirements for the drive bays and controller.
If you are sizing a UPS, measure the machine rather than adding up datasheet numbers. A Kill A Watt P3 P4400 plugged in ahead of the NAS gives a real wall figure for idle, working and startup draw, which is the input a UPS runtime chart actually needs.
Reasonable first purchases
For a single-drive backup target, the Synology DS124 is a one-bay unit with 1 GB of non-ECC DDR4 and a single gigabit port. Treat it as one copy in a backup plan, never the only copy, because a single drive has no redundancy by definition.
For documents, photos and a modest household file server, the Synology DS223 is a two-bay unit with 2 GB of fixed non-ECC DDR4 and a single gigabit port, drawing a quoted 17.3 W during access and 4.1 W with the drives hibernating. Fitted as a mirror it gives one drive of usable capacity with tolerance for one drive failure. Raw capacity is whatever two supported drives come to, so check the current drive compatibility list rather than assuming a ceiling.
It is a 2023 model, and the newer DS225+ is worth the comparison precisely because it moves the two limits this article keeps returning to. Its 2 GB of DDR4 stays onboard, but a single SODIMM slot takes the machine to 6 GB (2 GB + 4 GB), and it adds a 2.5GbE port alongside the gigabit one. If the plan is file sharing and backup, the DS223 is the cheaper answer and stays correct. If containers or a photo index are anywhere in the plan, the upgradable memory is the difference between a machine you can rescue and one you replace.
For four bays and up, compare an appliance against a build honestly. Once you are buying a four-bay enclosure and four NAS drives, the price gap narrows and the build starts to buy real things: more memory, ECC support, faster networking, and a chassis you can still expand in five years.
A short pre-purchase checklist
- Which of the three jobs is this machine for, and is anything on the shortlist specified for a different one?
- How much usable capacity remains after redundancy, and does that still meet the target?
- Are all candidate drives CMR, and does the workload rating match a continuously powered machine?
- Is the memory upgradable, and does the planned software fit in what ships?
- What is the network port, and is that ceiling acceptable for the largest transfer you care about?
- Does the enclosure have bays you are not filling yet, and can the power supply start every drive at once?
Spending a little more on bays and a little less on the processor is the version of this decision that ages well.
