What the cost readout measures
This page is an explanation. To actually make things lighter, see Automatic Optimization On Build.
In one line
The cost readout in the inspector is an estimate of the relative cost of the baked material after upload. It is a model estimate with the heaviest reference profile as 100%, not actual GPU utilization. Manager's Current load on the Start tab is the average of MingToon materials. It recalculates 10 seconds after edits stop and is not shown when multiple characters are selected.
What this number can and cannot do
| What it can do | What it cannot do |
|---|---|
| Compare which material is heavier within the same avatar | Convert into how many times longer a frame takes |
| Check whether the estimated cost dropped after turning a module off | Stand in for GPU milliseconds or SetPass counts |
| Compare estimated texture memory between materials | Judge the whole avatar's cost from the highest value |
The model is tuned to one specific Built-in benchmark environment. Screen coverage, overdraw, light count, GPU and render pipeline all change the real cost.
Reading the cost panel
Open Cost Details from the load menu at the top of the Full Inspector. The details contain a reference profile, two estimate cards, and an upload breakdown table.
| Readout | How to read it |
|---|---|
| After upload estimate | Average estimated baked cost of selected materials |
| While editing estimate | Average estimated cost of selected editable materials |
| Upload breakdown · peak selected material | Module contributions of the heaviest selected material |
| Module share of upload total | Share of total upload cost including the base shader and corrections |
The model calculates the authoring state and the estimated baked state separately. Instruction count, always-executed texture reads, register usage and pass count go into it.
For layers it looks at the number actually used and the execution conditions, not the slot count. Ten slots does not make it count ten reads per pixel. Conversely, a strength of 0 does not make the code disappear. Values that can change at runtime are kept.
What that constrains
- The scene-wide cost of making camera depth available is not part of the material multiplier. Compare the change in a real scene the first time you turn a depth effect on.
- Color correction driven by animation or a script has to keep working after the bake, so it does not fold into a constant. That much estimated cost stays.
- A lower estimated grade alone does not confirm an improvement. Compare real builds at the same camera, distance and lighting.