hostinglinks.
PRACTICAL TECH, EXPLAINED

SYSTEM MEMORY

How much RAM does a home server need?

Build a budget for the services that run together, then check it during a busy period. Buying RAM from a list of application names alone misses the difference between a quiet file server and one importing thousands of photos.

Updated 17 October 2026. The figures below are an illustrative planning exercise, not measurements of named applications. How we distinguish estimates from tests.

Make a budget before choosing 8, 16 or 32 GiB

List the operating system, applications, databases and any virtual machines. Record each application's published requirements, then allow for jobs that overlap: photo indexing, document OCR, backups and video conversion. A minimum requirement for one application is not a budget for the whole server.

Example budget for a small Linux server, in GiB
Budget itemAssumed allowance
Operating system and background services2
Applications and their databases together4
Additional overlapping import or backup work2
Spare capacity for variation and growth2
Total planning budget10

For these assumptions, 8 GiB falls short and 16 GiB leaves 6 GiB beyond the budget. There is no demonstrated reason to choose 32 GiB yet. If you also reserve 8 GiB for a virtual machine, the budget becomes 18 GiB and 16 GiB no longer covers it. Replace every allowance with evidence from your own workload before buying.

Do not add the entire observed peak to an application allowance that already includes it. Equally, summing idle readings can underestimate demand. On a virtualisation host, count guest allocations and host overhead; do not count the same guest applications twice.

Measure a real busy period on Linux

  1. Run free -h before starting work. Note the available memory and swap figures.
  2. Start a representative import, backup or indexing job while the normal services remain running.
  3. Watch free -h again and use vmstat 1 for a short live sample. After its first report, the si and so columns help show current swap activity. Stop with Ctrl+C.
  4. Record the lowest available memory, sustained swapping, failed jobs and responsiveness. Repeat with the combinations you actually expect to run together.

Linux uses spare RAM for caching. A small “free” figure alone does not prove that an upgrade is needed. The kernel's MemAvailable estimates what could be used by new applications without swapping. Swap already occupied after an earlier busy period is also different from continuous swapping during your test.

Check Docker without mistaking its limit for usage

Use docker stats to watch running containers, or take a single snapshot:

docker stats --no-stream

Read both sides of MEM USAGE / LIMIT. A container showing a large limit has not necessarily consumed that amount. Docker's Linux display subtracts some cache from usage, so adding its figures will not reproduce total host memory use. Keep the host reading alongside the container readings. A one-off snapshot can miss a short peak.

When to upgrade, and when to reschedule

If overlapping jobs cause pressure, try scheduling the bulk import outside the backup window and measure again. If normal daily work still exhausts available memory, repeatedly swaps or loses processes, more RAM may be justified. Container limits can stop one service taking everything, but setting them too low can make that service fail.

Check the exact machine's maximum memory, spare slots and whether its RAM is soldered before purchasing. Local AI needs a separate calculation: ordinary RAM does not become dedicated GPU VRAM. Use the local AI memory calculator for that workload.

Technical references