Add the single numbers
Memory adds up. Eight stacks measured alone summed to 1,113 MiB; together they rose by 1,101 MiB. Caddy for HTTPS adds about 6 MiB.
Flare
2 GB VPS hosting · Flare
Flare is our 2 GB plan: 1 vCPU, 25 GB SSD, unmetered transfer. We ran real stacks on it and measured idle memory, so you can budget before you buy.
Flare from $5.99/month · USD · availability varies by region
Overview
One Flare server, Docker, and the stacks from our own guides: each alone, then together, then with a swap file. Every figure is idle. Treat it as a floor, not a promise.
Uptime Kuma, Vaultwarden, Navidrome, Audiobookshelf, Forgejo, Jellyfin, Home Assistant and n8n ran together at idle. They added about 1.1 GB and left about 425 MB free, with no swap.
Immich alone took available memory down to 38 MB in about 35 seconds. Plausible only fit once we added a 2 GB swap file. Swap is a safety net, not more RAM.
Nine sets of images take 12.7 GiB of the 25 GB disk, and Ubuntu with Docker holds 2.3 GiB. That leaves about 8.5 GiB for data, backups and logs.
Workflow
Add up what you plan to run, then check it on the server.
Memory adds up. Eight stacks measured alone summed to 1,113 MiB; together they rose by 1,101 MiB. Caddy for HTTPS adds about 6 MiB.
Idle is a floor. Logins, imported data, libraries and traffic cost more. Keep headroom and watch free -m.
If available memory stays low or the swap file is busy, pick a bigger plan instead of leaning harder on swap.
The measurement figures on this page come from our published post. It has the full tables and the method.
One Flare VPS with Docker. Each stack started clean three times. Idle RAM is the rise in used memory after two minutes. A control container showed the method resolves roughly plus or minus 40 MiB.
Caddy and HTTPS (they need a real domain), accounts, logins, imported data and load tests. The Jellyfin, Navidrome and Audiobookshelf libraries were empty.
A 2 GB swap file at the default swappiness of 60. Running swapoff with all nine stacks up made the kernel end the swapoff process. Stop stacks before removing swap.
Pricing
Monthly pricing in USD. Unmetered transfer, subject to the acceptable use policy.
This plan
Small projects & personal apps
$5.99/ month
Billed monthly. No setup fee.
Deploy FlareWebsites & lightweight services
$7.99/ month
Billed monthly. No setup fee.
Deploy OrbitGrowing applications
$15.99/ month
Billed monthly. No setup fee.
Deploy CometMore demanding workloads
$27.99/ month
Billed monthly. No setup fee.
Deploy PulsarHeavy, memory-hungry workloads
$49.99/ month
Billed monthly. No setup fee.
Deploy QuasarOur largest plan
$89.99/ month
Billed monthly. No setup fee.
Deploy NebulaIncluded
Always on. No scrubbing fee, no attack-time surcharge.
Automated, seven-day retention, restore them yourself.
Save a point to return to. Up to 7, free.
One public IPv4 of your own.
Console access from your browser, even without SSH.
No per-GB charge, subject to the Acceptable Use Policy.
Your public key is on the machine at first boot. No password sent by email.
Pay in crypto at checkout, not only by card.
Servers, balance, invoices and tickets from your phone.
Answers
Yes. The measurements ran on Docker. With Docker installed and nothing running, the host showed about 441 MiB used.
No. On its own, without swap, it took available memory down to 38 MB and we stopped it first. Our Immich guide recommends the 8 GB Comet plan.
Immich, or anything with a machine-learning container. Plausible beside other apps. More than two of Home Assistant, n8n and Plausible, and only with swap. Transcoding and compiling were not measured.
Our guides call Flare the fit for Vaultwarden, Forgejo, WireGuard with Pi-hole or AdGuard Home, Headscale, SearXNG and Discord or Telegram bots. They also say Flare cannot hold a Valheim server, and that 2 GB is fine for Minecraft only with two or three players on a small map.
Unmetered means no per-GB transfer charge. It does not mean unlimited throughput, and the acceptable use policy applies.
Check your region and current stock in checkout, then measure your own workload.
Deploy Flare