CT 161 gamelab is an unprivileged LXC container on my Proxmox host: Debian 12.15, four cores of an Intel Core i7-4790K at 4.00 GHz, 8,192 MB RAM, 2,048 MB swap. I installed the Project Zomboid dedicated server there on 2026-10-18 using LinuxGSM v26.4.0. ./pzserver auto-install logged Success! App '380870' fully installed, and the server version was 42.21.0.
The box is LAN only. Nobody has ever joined this server. A minute after ./pzserver start, the ProjectZomboid64 process was holding 7,873 MB. When the run ended, free -m showed 128 MB free and 1,400 MB in swap. The process jumped from 2,202 MB to 7,873 MB between 31 and 61 seconds, before ready at 92 seconds.
Where the 8 GB cap comes from
serverfiles/ProjectZomboid64.json ships with this block:
"vmArgs": [
"-Djava.awt.headless=true",
"-Xmx8g",
"-Dzomboid.steam=1",
"-Dzomboid.znetlog=1",
"-Djava.library.path=linux64/",
"-Djava.security.egd=file:/dev/urandom",
"-XX:+UseZGC",
"-XX:-OmitStackTraceInFastThrow"
]-Xmx8g is the maximum Java heap. -XX:+UseZGC selects the ZGC garbage collector. I changed only the -Xmx value for the second run. Everything else stayed as shipped.
The measurement setup
I measured both starts with gmeasure.py on the container. It cold-starts ./pzserver start, polls /proc/<pid>/status every second, and marks ready when the console log prints *** SERVER STARTED ****. RSS and process swap come from VmRSS and VmSwap. CPU was sampled from /proc/<pid>/stat over a 60-second window after the ten minutes of RSS sampling; the window measured 3.7% of one core at -Xmx8g and 3.9% at -Xmx4g. Zero players joined either run.
The shipped heap did this
Run 1 kept -Xmx8g. Start time was 16:44:57. At 31 seconds the process RSS was 2,202 MB and swap was 0 MB. At 61 seconds it was 7,873 MB with 97 MB of process swap. Peak RSS came in at 7,891 MB. Final RSS after ten minutes was 7,617 MB, with a process tree at 7,621 MB.
free -m at the end showed 7,861 MB used, 128 MB free, 330 MB available, and 1,400 MB in container swap.
What the counters reported
An earlier start at 16:44:00 with the same settings left free -m at 7,937 MB used, 67 MB free, and 1,529 MB swap. The ProjectZomboid64 process reported RSS 7,968,480 kB, VmSwap 200,468 kB, RssAnon 908,524 kB, RssFile 33,776 kB, and RssShmem 7,026,180 kB.
Of the roughly 7.6 GB resident, 7.0 GB was counted as shared memory and 0.9 GB as anonymous memory. The container’s swap use was 1,400–1,529 MB, far above the process’s own VmSwap of 156–200 MB. Those are the numbers the captures show. They do not line up, and I left it there.
One line changed
At 16:56:12 I changed only -Xmx8g to -Xmx4g in ProjectZomboid64.json and started again. At 31 seconds RSS was 2,089 MB with no swap. At 61 seconds it was 4,864 MB, still no swap. Ready came at 69 seconds. The server settled at 5,176 MB by the end, with peak RSS 5,190 MB and final RSS 5,142 MB.
The container’s free -m after the run showed 5,144 MB used, 2,118 MB free, 3,047 MB available, and 9 MB in swap. At a 4 GB heap cap, the process still used 5,142–5,190 MB, roughly 1.1 GB above the cap. Each configuration was run once. I did not repeat them.
Side by side
| heap cap | ready | rss peak | rss final | container swap |
|---|---|---|---|---|
-Xmx8g | 92 s | 7,891 MB | 7,617 MB | 1,400 MB |
-Xmx4g | 69 s | 5,190 MB | 5,142 MB | 9 MB |
The VPS arithmetic
Both runs are rows in the server requirements table, and the number that actually decided the plan was in the JSON file.
The measured 5.1 GB at a 4 GB cap does not fit a 4 GB plan. It fits an 8 GB plan. The shipped 8 GB cap filled an 8 GB box into swap at zero players, so leaving it alone means either a 16 GB plan or a config edit before first start.
plans.py reads the same price file as the hosting cost calculator. These are list prices from each provider’s public pricing page, checked 2026-10-18.
| Provider | Plan | RAM | vCPU | Price |
|---|---|---|---|---|
| Hetzner | CPX32 | 8 GB | 4 vCPU | EUR 35.99/mo |
| Vultr | Cloud Compute Regular 8 GB | 8 GB | 4 vCPU | USD 40.00/mo |
| DigitalOcean | Basic Regular 8 GiB | 8 GB | 4 vCPU | USD 48.00/mo |
| Provider | Plan | RAM | vCPU | Price |
|---|---|---|---|---|
| Hetzner | CPX42 | 16 GB | 8 vCPU | EUR 69.99/mo |
| Vultr | Cloud Compute Regular 16 GB | 16 GB | 6 vCPU | USD 80.00/mo |
| DigitalOcean | Basic Regular 16 GiB | 16 GB | 8 vCPU | USD 96.00/mo |
Hetzner shows a “from” price in EUR. At the time of writing, those are the figures.
The update catch
ProjectZomboid64.json lives in serverfiles/, the directory SteamCMD installs and updates. I have not run a SteamCMD update since editing it, so whether an update restores -Xmx8g is not observed. It is worth checking after every update before you start the server.
Same box, other games
Same container, zero players, one game at a time. Valheim used 1,432 MB. Vanilla DayZ used 4,842 MB. Project Zomboid used more than either at both heap caps.
Before you rent a VPS for a Project Zomboid server, open serverfiles/ProjectZomboid64.json and set the heap cap to fit the plan you are about to pay for. On an 8 GB plan, the run here says -Xmx4g: 5.1 GB used, 9 MB in swap, and the server came up 23 seconds sooner.
