Unity Performance Audit and Frame-Time Diagnosis
Your game can average 60 FPS and still feel terrible. We investigate worst-frame behavior on the actual target hardware, distinguish CPU scheduling from GPU work, and turn profiler evidence into a production-ready repair plan.
When a performance audit is the right first purchase
Teams often arrive with a list of suspected optimizations: batch more meshes, replace the grass system, rewrite AI, or disable volumetrics. The problem is that the list is not evidence. An expensive main-thread wait marker might be caused by a GPU fence; a camera-turn hitch might be a shader warm-up rather than draw calls. Fixing the wrong subsystem can consume an entire sprint with no effect on the slowest frames.
We begin by defining the target: device, scene, resolution, capture route, minimum frame rate, loading transitions and visual constraints. For a 60 Hz target, the total frame budget is about 16.67 ms, but that does not mean simply adding the CPU and GPU durations. The limiting path depends on overlap, scheduling and presentation.
Critical camera angles
A stadium or combat arena may run well from one camera but stall when crowds, shadow casters and transparency fill the frame.
Event-driven hitches
Weapon fire, respawns, UI, replays, save writes and first-time effects require dedicated captures, not only an idle benchmark.
Long-run degradation
A ten-minute session can miss asset retention, memory pressure or gradual frame pacing deterioration.
Platform divergence
The same Unity scene can bottleneck for different reasons across Steam Deck, desktop GPUs and consoles.
Capture protocol, not guesswork
- Freeze the reproduction: Unity version, scripting backend, build flags, pipeline, exact scene route, quality preset, render scale and camera movement.
- Collect CPU Main Thread and Render Thread timelines, GC.Alloc markers, memory counters and platform GPU timings where accessible.
- Separate cold-start, warm-steady-state, worst-combat and transition recordings. Captures of only the Editor are insufficient.
- Use multiple runs to compare median, p95/p99 frame time and hitch frequency; record the exact methodology so it can be repeated.
- Design a reversible A/B experiment that targets one suspected subsystem, then validate in a comparable player build.
We distinguish time spent executing work from time spent waiting for another producer. A marker such as Gfx.WaitForPresentOnGfxThread can be a symptom of presentation or GPU pacing, not proof that render-thread CPU instructions are expensive. Deep Profiling also changes the measured workload, so detailed captures and final production-like verification play different roles.
Typical bottleneck triage matrix
| Observed signature | Evidence to collect | Likely next action |
|---|---|---|
| Regular spikes at a fixed interval | GC.Alloc, collector markers, recurring code path | Remove high-frequency allocations, verify retained object lifetime |
| GPU-bound at high resolution | Per-pass GPU times and resolution scale sweep | Test pixel-heavy passes and target-platform quality tiers |
| CPU frame grows with agents | Animation, navigation, physics, script update markers | Fix scheduling and visibility update strategy |
| First encounter or first particle hitch | Cold versus warm run, shader/asset initialization | Target controlled prewarm or streaming sequence |
| Hitch only after menu or scene load | Async operation timing, disk IO, main-thread work | Rework load boundaries and contention rather than raw render cost |
The report should make confidence explicit: reproduced and measured, suspected but unverified, or blocked by missing instrumentation. We do not collapse those categories into a list of generic tips.
What you receive
Capture manifest
A record of device, scene route, build hash and profiling configuration so the measurements can be reproduced by your team.
Bottleneck register
The top issues ranked by player impact, estimated fix risk, dependency and confidence in the diagnosis.
Before/after protocol
A repeatable test definition and acceptance criteria for each proposed fix, not an unexplained FPS screenshot.
Delivery roadmap
A practical order for high-confidence corrections, investigations requiring more data and architectural work needing a larger milestone.
An audit is distinct from a full optimization implementation. If you want us to carry out the fixes, we can agree a second milestone based on what the measurements actually reveal.
How this connects to production visuals
Rendering optimizations cannot be evaluated independently of the game's presentation. A replay camera needs believable shadows, a hero weapon needs high-fidelity surface response, and an outdoor environment needs sufficient depth to preserve visibility. Our background in Unity presentation and optimization makes those visual constraints part of the acceptance review rather than an afterthought.
See our Lacrosse 26 contribution for a real example of the kind of scene-presentation work Fyros has performed. This page does not claim a specific measured improvement from that project.
What we need from your team
Provide a player build or source access, target device list, a repeatable performance complaint, the frame-rate and image-quality requirements, and the next release milestone. If you already have captures, include them even if they are incomplete.
For private console hardware or SDK-dependent problems, all access remains subject to the platform holder's developer program and the client's authorization. The first conversation establishes whether a review is viable and what access is required.
Engagement and acceptance criteria
A focused engagement starts with a problem statement and ends with work your team can test. We agree on scope, deliverables, review milestones, data access and acceptance criteria before implementation. Outcome depends on the state of the project and measured constraints.
Questions production teams ask
Can you audit an existing Unity project without rewriting it?
Yes. The diagnostic milestone is scoped around evidence and a fix plan. Code changes are a separately agreed implementation scope.
Does a performance audit include PS5 or Xbox testing?
Only when target hardware, approved project access and platform-specific tooling are available and included in the scope.
Will you guarantee 60 FPS?
Not before seeing the build. We can define measurable acceptance criteria after baseline captures identify what is technically feasible.
Is a profiler screenshot enough?
Usually not. Reproduction steps, target platform, timing distributions and associated capture data are more useful.
Technical references and further reading
Relevant vendor documentation for validating version-specific behavior. These links are references, not claims of a measured result from an undisclosed project.
Frame drop root-cause investigation | General Unity optimization service
Explore other specialist Unity engagements or learn about co-development for active productions.
Get the problem fixed, not just diagnosed
Fyros Games Studio works with studios and publishers on scoped Unity engineering, rendering, optimization and production tasks. Share the technical blocker and we will discuss a measurable work package.
TALK TO FYROS