title: Campfire vs Slirc
crumb: Reports / Campfire vs Slirc
description: Measured resource comparison: Campfire (Rails + Redis) against Slirc (Bun + Ergo IRC), RAM, CPU, processes, and image size.
lead: Same job, group chat on our own hardware. Very different footprint.

Campfire is an excellent alternative: open source, self-hosted, runs one command, and [you get the code](https://once.com/campfire). We measured both products side by side to see what the same job costs on our own hardware.

## The numbers

| metric | Campfire | Slirc |
| --- | --- | --- |
| RAM, idle | 1.62 GB (76% of its own 2 GB cap) | ~20 MB total (13.6 MB app + 6.6 MB Ergo) |
| CPU, idle | 2.03% | ~0.1% |
| processes | 220 | 12 |
| image size | ~917 MB (877 MB app + 40 MB Redis) | ~135 MB (97.8 MB app + 36.7 MB Ergo) |
| extra services | separate Redis container | none |

## How it adds up

- RAM: **81×**
- CPU: **~20×**
- processes: **18×**
- disk images: **6.8×**

## Capacity

Campfire sits near its own 2 GiB cap before anyone connects. Slirc's whole stack uses about 1% of that, so the same cap holds roughly 80 Slirc stacks.

Campfire alone eats 1.6 GB of its machine's 31 GB. Slirc's whole stack fits comfortably on its 3.8 GB host.

## Load

Synthesized 250 concurrent IRC clients against the local Ergo instance: all registered and joined, then chattered at ~210 messages/s for 40 seconds, about 52k outbound deliveries after fanout.

| metric | 250 users (measured) | 500 users (modeled) |
| --- | --- | --- |
| Ergo RAM, idle | 17 MiB | 17 MiB |
| Ergo RAM, during load | 39 MiB | 60-80 MiB |
| Ergo CPU, sustained | 5-15% of one core | ~10-30% |
| Ergo CPU, burst spike | ~49% of one core | under one core |
| messages handled | 8.4k in / 52k out | ~2× |

This is the chatty worst case, every user talking continuously. Real groups sit mostly idle at the WebSocket relay, so actual capacity is well above the model.

## Recommended server

Same ladder shape Campfire publishes (they start at 250 users / 2 GB / 1 CPU). Slirc's numbers start from the measured load above and scale for the whole stack, app relay, Ergo, Caddy, and headroom, not bare Ergo. [Hetzner Cloud](https://www.hetzner.com/cloud/) ARM servers hit these tiers out of the box, prices included.

| concurrent users | RAM | CPU | Hetzner | est. price |
| --- | --- | --- | --- | --- |
| 250 | 1 GB | 1 CPU | cax11 | ~€7/mo |
| 1,000 | 2 GB | 2 CPU | cax11 | ~€7/mo |
| 5,000 | 4 GB | 4 CPU | cax21 | ~€13/mo |
| 10,000 | 8 GB | 8 CPU | cax31 | ~€25/mo |

Memory stays the cheap axis (Ergo idles at ~17 MiB and peaks under 40 MiB for 250 chatty users); the growth curve is CPU and the WebSocket fanout, which is why it climbs slower than Campfire's ladder.

## Method

Both hosts measured 2026-09-07 with `podman stats` (three samples, two seconds apart), `podman images`, and `podman system df`. Load numbers measured 2026-09-15 with a 250-client synthetic script against the local stack.