How We Measure Server Boot Speed

We published a benchmark: Zeros Host servers cold-boot in around 6 seconds. Vanilla GMod hosting takes 45-90 seconds. We call that around 7x faster.

That is a meaningful number. Meaningful numbers invite scepticism, and scepticism is reasonable. This post explains what we actually measured, where the vanilla figure comes from, and why we use the conservative number.

What "cold boot" means

A cold boot is the time from sending a start command to the server being reachable by players. Specifically, we time from the moment the panel dispatches the start signal to the first successful A2S_INFO response, the handshake a game client uses to confirm a server is online and accepting connections.

We do not time from when the process appears in Docker. We do not stop the clock at an internal log line that happens before the server is actually joinable. The end marker is a real external probe: the same packet your game client sends when you browse the server list.

The scenario is sandbox/gm_construct with no addons and no workshop content. That is the lightest possible configuration and the fairest comparison to a fresh vanilla install.

How we measure it

We use a scripted measurement that runs five cold starts and records phase timestamps from the server console. Each run is cold, with pagecache dropped before the start. The script reports mean and standard deviation.

Our canonical cold-start result is 6.2 seconds mean (n=5, production hardware, 3 June 2026). The full methodology and raw numbers are on the /benchmarks page. We publish the standard deviation alongside the mean because a mean without spread is not a real benchmark.

Where the 45-90 second vanilla range comes from

We do not run vanilla GMod hosting on a competitor's infrastructure. We cannot. The 45-90 second range comes from community server setup guides, Facepunch forum threads, and Steam Community Hub discussions where server operators document how long their servers take to come online.

This is not a controlled benchmark. It is a reported range from a large number of independent operators running a variety of hardware. The low end (45s) appears in guides for dedicated servers on good hardware. The high end (90s) appears in reports from shared hosting and slower VPS setups.

We are transparent about this on the /benchmarks page. The vanilla figure is community-sourced, not something we measured ourselves.

Why ~7x and not higher

The arithmetic supports a higher figure. 6.2 seconds against a 60-second midpoint is roughly 10x. Against the 90-second ceiling it is nearly 15x.

We use ~7x because it is the most defensible number. It holds against the low end of the vanilla range (45 seconds divided by 6.2 seconds = 7.3x). If someone runs a well-tuned vanilla server and gets 44 seconds, we are not going to argue. The claim survives that.

We do not use the ceiling figure because we did not measure the ceiling. We use the floor because that is the honest test of whether the claim holds.

What we do not claim

We do not claim 7x for DarkRP servers with large addon sets. A DarkRP server loading thousands of Lua files takes significantly longer to cold-boot in vanilla: the measured community range for that configuration is wider and the comparison ratio is higher. We do not use that number in the hero claim because the baseline scenario (sandbox, no addons) is the controlled one.

We do not claim that boot speed is the only thing that matters. There are hosting decisions that have nothing to do with startup time. We benchmark what we can measure. The full archive covers game server storage, panel performance, and website load times as well as boot speed.

The open benchmarks commitment

Every number on the /benchmarks page is either verified by a script running against production hardware or marked as documented from a PR description. Negative results (attempts that did not work) are published alongside wins. We have shipped several optimisation rounds that moved nothing, and those are on the page too.

If a future change improves boot speed further, the page will be updated with the new number and the new raw data. If a change makes something worse, that goes on the page too. The point is that the numbers are not marketing copy that lives in a slide deck. They are measurements that live in a version-controlled data file, reviewable by anyone.

Continue reading

This post is part of the Fast Load series. Related articles: Fast Load: How We Are Cutting GMod Boot Time by 87%, How We Are Building Sub-Second Cold Start. The Fast Load feature page shows live benchmarks and what the feature means for your server.