Start by measuring the failure. Low client FPS, high resource time, server hitches, database waits and network loss are different problems. Capture a repeatable baseline at a representative player count before removing resources or copying convars from another server.
Method: reviewed 9 August 2026 against the official Cfx.re profiler guide, client console commands Ve server commands.
1. Define the symptom and baseline
- One player’s FPS drops: reproduce on that client and inspect the client resource monitor.
- All players hitch: record server hitch warnings, resource timing, database latency and host CPU saturation.
- Joining is slow: inspect streamed asset size, download behaviour and initialization work.
- Actions lag: trace the event and database query rather than blaming graphics.
Record artifact version, player count, active resources, test route and timestamps. Change one variable at a time.
2. Use resmon and the profiler correctly
Open F8 and use resmon 1 for a quick client-side comparison. Treat it as a pointer, not a verdict: timing varies with what the player is doing. Reproduce the same interaction several times. For code-level evidence, take a profiler capture during the slow action and inspect the expensive scope or event.
Keep the detailed FiveM resmon guide open while testing.
3. Fix resource code at the bottleneck
- Replace unconditional per-frame loops with event-driven work or a justified wait interval.
- Validate network events on the server and send only the data the client needs.
- Avoid broadcasting large payloads when a targeted event is sufficient.
- Cache stable lookups, but invalidate the cache when the underlying state changes.
- Profile before and after; shorter code is not automatically faster code.
4. Check database work
Log slow queries with parameters and call sites. Add indexes only after confirming the query pattern, and avoid repeated queries inside loops. Batch writes when correctness allows it. Test reconnects, scheduled saves and peak economy actions because an empty staging database can hide production latency.
5. Budget streamed assets
Audit oversized textures, duplicate models and resource packs that stream far more than the server uses. Test a clean join and a busy location. Verify levels of detail, texture resolution and file integrity before compressing or removing assets. A broken upload can look like a performance problem.
6. Separate host limits from script limits
Watch per-core CPU saturation, memory pressure, storage latency, packet loss and database location. FiveM workloads can bottleneck one busy core while total CPU still looks low. Compare hosts with the same build and load; RAM and advertised slots alone do not prove capacity.
7. Roll out and monitor
- Back up the resource, configuration and affected database tables.
- Deploy one measured fix to staging.
- Repeat the baseline scenario and compare the same metrics.
- Release during a controlled window and watch hitch, error and player reports.
- Roll back immediately if the target metric or gameplay flow regresses.
Optimization is complete only when the measured symptom improves without moving the failure elsewhere.