← All Articles

Mini PC RAM Upgrades for a Homelab: What to Buy and What Actually Fits

RAM is the cheapest upgrade that makes a homelab mini PC useful. Here's how much you actually need, what your machine will really accept, and which kits to buy without wasting money.

hardwaremini-pcupgradescost-analysis

This post contains affiliate links. If you buy through them, I earn a small commission at no extra cost to you.

The single best value upgrade in a budget homelab is not a faster CPU or a bigger SSD. It is RAM, and it is not close.

A $130 mini PC with 16GB will run a Docker stack that keeps most people happy for a year. The same machine with 8GB will have you killing containers to start other containers within a week. The gap between those two experiences costs about $30 and takes ten minutes with a screwdriver.

The problem is that the buying advice is a mess. Vendor spec sheets understate what boards accept, Intel’s official maximums are conservative to the point of being wrong, and half of the “will this work” answers on Reddit are about a BIOS revision from two years ago. Here is how to get this right the first time.

How much RAM a homelab actually needs

Start from what you plan to run, not from a round number. Real idle footprints on a typical setup:

What you’re runningRealistic RAM use
Proxmox host itself (no guests)1.5 to 2.5GB
Docker engine plus 10 small containers3 to 5GB
Jellyfin or Plex (idle, no transcoding)400MB to 1GB
Immich with machine learning enabled2 to 4GB
Home Assistant OS as a VM2GB allocated
Nextcloud AIO2 to 3GB
A Postgres or MariaDB container300MB to 1GB
ZFS ARC (if you use ZFS)Will happily take 50% of the box

The pattern most people hit: 8GB is genuinely tight the moment you add anything with a database. 16GB comfortably runs a dozen containers and leaves headroom. 32GB is where you stop doing mental arithmetic before installing something, and it is the right target if you run Proxmox with VMs as well as LXC containers, because a VM reserves its allocation whether it uses it or not.

If you are running ZFS, add memory on top of all of this. ARC is not optional overhead, it is what makes ZFS fast, and starving it is a common self-inflicted performance problem.

My rule: buy 32GB unless you are certain you’ll stay container-only forever. The price gap between a 16GB and a 32GB kit is smaller than the cost of buying twice.

Check what your machine takes before you buy anything

Three things decide what you can install, and you can find all of them without opening the case.

On Linux, including a Proxmox host:

sudo dmidecode -t memory | grep -E "Size|Type:|Speed|Form Factor|Locator"

That tells you the module type (DDR4 or DDR5), the current speed, the form factor (SO-DIMM on essentially every mini PC), and critically, how many slots exist and how many are populated. A machine reporting two slots with one populated is an easy upgrade. A machine reporting one slot means any upgrade is a replacement, and the stick you pull out is wasted money.

Then check the physical constraint:

What the common homelab mini PCs actually accept

Vendors revise these silently mid-production run, so treat this as a starting point and confirm against your specific unit with dmidecode.

Machine classTypeSlotsOfficial maxCommonly working
Intel N100 / N150 boxes (Beelink S12 Pro, most Topton units)DDR4-3200 SO-DIMM116GB32GB single stick, BIOS dependent
Beelink EQ12 / EQ13 classDDR5-4800 SO-DIMM116GB32GB single stick, BIOS dependent
Ryzen 7000 / 8000 series (MINISFORUM UM790 Pro, Beelink SER8)DDR5-5600 SO-DIMM264GB96GB with two 48GB modules
Used ThinkCentre / OptiPlex Tiny, 8th to 10th gen IntelDDR4-2666 SO-DIMM232 to 64GBMatches official, generally well behaved

The N-series situation deserves a direct answer because it comes up constantly. Intel’s official maximum for the N100 is 16GB. That number is a platform validation limit, not a hard controller limit, and a single 32GB DDR4 or DDR5 SO-DIMM works on a large share of N100 boards. It is not universal. Some BIOS versions post fine, some hang, some boot but report 16GB. Buy from a seller with a painless return window and you have a cheap shot at doubling your capacity.

Note that every N-series chip here is single channel regardless of how many slots the board has. You cannot fix that with RAM. It also means there is zero performance argument for buying a matched pair on those machines, which simplifies the decision.

What to buy

For homelab use, three things matter and one thing does not.

Capacity matters. Buy the most you can use.

Voltage and standard matter. DDR4 SO-DIMM at 3200MHz, or DDR5 SO-DIMM at 4800 or 5600MHz, matching what your board specifies.

Brand matters a little. Crucial, Kingston, Samsung, and SK Hynix modules are the safe picks. Crucial is Micron’s own retail brand, so you are buying the chip manufacturer’s product directly, which is why it tends to be the most boring and reliable option. No-name Amazon-only brands are usually fine and occasionally are not, and the $8 you save is not worth a machine that fails to post.

Speed does not matter. Nobody’s Jellyfin library indexes faster on DDR5-5600 than DDR5-4800. Homelab workloads are bound by disk, network, and capacity. Overpaying for frequency on a homelab box is pure waste, and on a single-channel N100 the board will likely down-clock it anyway.

Reasonable starting points, priced as of writing and moving constantly:

UpgradeTypical priceGood for
32GB DDR5-5600 SO-DIMM, single$70 to $95Single-slot DDR5 mini PCs, the standard upgrade
32GB DDR4-3200 SO-DIMM, single$50 to $70N100 boxes and used ThinkCentre Tiny units
2x16GB DDR5-5600 kit$75 to $100Dual-slot Ryzen machines, gets you dual channel
2x32GB DDR5-5600 kit$150 to $200Ryzen boxes running a real VM workload
48GB DDR5 SO-DIMM modules$130 to $180 eachThe 96GB path on dual-slot Ryzen machines

Those 48GB modules are worth knowing about. Higher-density DDR5 chips made non-power-of-two capacities possible, so a two-slot machine can reach 96GB. Support is less universal than 32GB modules and it is an expensive experiment to get wrong, so check your specific model’s BIOS notes first.

Skip ECC, and skip the arguments about it

You will find threads insisting a homelab needs ECC memory, usually attached to a ZFS discussion. For the machines covered here it is a non-question: consumer mini PCs do not support true ECC, and no SO-DIMM you buy will change that.

DDR5 has on-die ECC, and it is not the same thing. It corrects errors within the memory chip itself and reports nothing to the operating system. It is a manufacturing yield feature, not a data integrity feature you can monitor or rely on.

If you genuinely need ECC, you are shopping for used enterprise hardware or a Xeon-D board, which is a different budget and a different power and noise profile than anything in this article. For a home Jellyfin, Immich, and Docker stack with real backups, non-ECC is what nearly everyone runs and it is fine.

Installing it

Ten minutes, one screwdriver, three things worth knowing.

  1. Unplug it and hold the power button for five seconds to discharge. Then touch the bare metal chassis before touching the module.
  2. SO-DIMMs go in at roughly a 30 degree angle, then press flat until the side clips snap. They need more force than feels correct. A module that is seated but not clipped will post intermittently and drive you insane.
  3. First boot after an upgrade can take 30 to 60 seconds with a black screen while memory trains. This is normal on DDR5 especially. Do not panic and power cycle mid-train.

If it does not post: reseat the module, try the other slot if there is one, and if you swapped a working stick for a bigger one, put the original back to confirm the machine still boots. That isolates a bad module from an unsupported capacity in about two minutes. Clearing CMOS is the next step, and after that you are likely looking at a capacity the BIOS will not accept.

Once it boots, confirm the system actually sees it with free -h and sudo dmidecode -t memory. A machine that posts but reports half the RAM you installed is telling you the board took the module and capped it.

When a RAM upgrade is the wrong answer

RAM fixes one problem: not enough memory. It does not fix a slow CPU, a saturated 1GbE link, or a SATA SSD bottlenecking a database.

Check htop before you buy. If you are sitting at 40% memory use and your CPU is pinned during Immich imports or Jellyfin transcodes, more RAM changes nothing. That is a compute problem, and the fix is a better machine, which is a different conversation covered in the mini PC buying guide and the Beelink versus MINISFORUM comparison.

The other case worth thinking about: a second node versus more RAM in the one you have. A used mini PC under $200 gets you a second machine, more total cores, and actual redundancy for maybe three times the price of a 32GB stick. If the goal is capacity, RAM wins on cost every time. If the goal is being able to reboot the host without your DNS going down, the second node is the better $200.

For a single-machine homelab that is out of memory, though, this is the highest return-per-dollar upgrade available, and it stays with you when the machine gets demoted to a backup node later. Buy the 32GB, spend ten minutes, stop thinking about it.

More on where hardware money actually goes: what a homelab really costs.