Rendering analysis for Unity

Stop guessing
what to optimize.

Find rendering hotspots. Understand what's behind them. GPUSight connects visual scene maps, object evidence and GPU tests so you can choose your next optimization with context.

Unity Editor extension Version 1.0.0

SCENE VIEW → INVESTIGATEGPUSight / Unity Editor
GPUSight in Unity with a geometry workload map over Viking Village and an overview of rendering evidence.
A real GPUSight capture. Open any screenshot to inspect the details.
Find the pattern in your scene. Follow the evidence.

See the big picture

Your scene.
A different point of view.

Find areas worth investigating directly in Scene View. Switch maps to reveal different aspects of the rendering workload, then follow a finding into the objects and materials behind it.

Geometry workload map over the Viking Village buildings, with high-contribution surfaces in orange and red.
Geometry workload

See how geometry workload is distributed across the scene. Follow highlighted contributors into their object and material evidence.

Workload signal
LOD coverage map showing the declared LODGroup membership of buildings and props.
LOD coverage

Find objects without an LODGroup and see how many levels are declared. Continue with LOD-level and reduction views to inspect transitions.

Scene facts
Texture-resource map with highlighted surfaces and material resource areas in Viking Village.
Texture resources

Find surfaces associated with large texture resources, then inspect the material bindings. Native allocation data is currently available on Metal.

Resource evidence
Shader-source complexity map across buildings and foliage in Viking Village.
Shader complexity

Locate materials with higher source-complexity indicators. Use the pattern to choose a shader to research and test.

Source estimate

Real captures from Viking Village. Camera positions differ between examples; these are different views, not a before-and-after comparison.

More ways to investigateCoverageOpaque & transparent overdrawShadow castingAvailable passesShader featuresTexel density

Break down the workload

From a hotspot
to the objects behind it.

Turn a visual finding into a focused investigation. Group the scene, identify the heavy contributors, and open the evidence for an individual object.

  • Group your way. Material, shader, pass, camera, scene, layer, tag or prefab.
  • Inspect in context. Geometry, LODs, material slots, shader settings and texture bindings.
  • Keep moving. Jump to scene selection, the Inspector or material research.

Rankings use available rendering signals to guide investigation, not exact GPU milliseconds per renderer.

BREAKDOWN / SCENE CONTRIBUTORSGPUSight / Unity Editor
GPUSight Breakdown view grouping scene contributors with object counts and geometry workload.
Group the workload, then inspect the contributing objects.

Investigate geometry and LODs

Give distant detail
a closer look.

Find missing LOD coverage and transitions that barely reduce geometry. See where your LOD setup deserves attention.

  • Explore LOD coverage and estimated levels.
  • Inspect triangle reduction between levels.
  • Connect LOD membership to contributor totals.
LOD coverage map alongside the GPUSight Breakdown view grouped by LOD membership.

Inspect textures and texel density

Make texture
density visible.

Spot inconsistent texture scale and unusual mip demand. Follow the result to the actual textures bound to a material.

  • Texel density, mip demand and outlier maps.
  • A visual texel grid for comparing surfaces.
  • Texture dimensions and available allocations.
The real GPUSight texel grid over a roof surface.Texture thumbnails, bindings, dimensions and native allocation values in GPUSight.

Research shaders and variants

Go beyond
“this shader is heavy.”

Explore what the shader source is doing. Inspect texture sampling, math, loops, branches and vertex processing, then turn the findings into a testable hypothesis.

  • Understand the source. Review static indicators with their analysis context.
  • Inspect variants. Compare selected configurations and keywords.
  • Go deeper. Explore function-level source metrics.

Source complexity is a heuristic. It is not a compiled GPU instruction count or a measurement of function execution time.

Put the hypothesis to a GPU test
RESEARCH / SHADER SOURCEGPUSight / Unity Editor
GPUSight shader research with source complexity, texture sampling, loops, branches and vertex-processing evidence.
Source-derived indicators help explain where to investigate next.
RESEARCH / SELECTED VARIANTGPUSight / Unity Editor
GPUSight selected shader variant analysis with stage metrics and function-level source evidence.
Explore the selected variant, then follow source evidence into individual functions.

Benchmark and compare materials

Put your next
change to the test.

Compare material configurations in a controlled preview. Keep the conditions visible, preserve a baseline, and investigate how the workload scales.

Which material configuration
is worth taking further?

Review the differences between A and B, then compare their GPU timings under the same preview conditions.

  • Inspect shader, keyword, property and texture differences.
  • Read medians together with sample ranges and backend.
  • Check the promising change in your real scene.

Benchmarks require active URP and a supported timing backend. Results belong to the controlled preview workload.

BENCHMARK / MATERIAL A/BGPUSight / Unity Editor
A completed GPUSight A/B material comparison with GPU timings, device, test conditions and sample ranges.
A real completed preview run. The numbers describe this capture, not a guaranteed speedup.

See what changes
when the workload grows.

Run grouped scenarios that vary resolution, draw count and geometry. Use the results to investigate what your material's preview workload is sensitive to.

  • Compare resolution, draw-call and geometry scenarios.
  • Read each median alongside its sample range.
  • Repeat noisy runs before drawing a conclusion.

This recorded run includes variability. Scaling tests support investigation; they do not automatically diagnose the bottleneck of your scene.

BENCHMARK / TEST SCALINGGPUSight / Unity Editor
GPUSight Test scaling results for eight resolution, draw-call and geometry conditions.
Eight grouped scenarios with visible conditions and sample ranges.

Give your next test
a useful reference.

Preserve a completed material benchmark as a baseline. After a change, compare compatible runs and export the report as JSON.

  • Keep the reference result and its test conditions.
  • See when conditions prevent a meaningful comparison.
  • Export samples and context for review.

A material baseline is separate from a scene snapshot. Both retain evidence and its original context.

BENCHMARK / BASELINEGPUSight / Unity Editor
GPUSight GPU time result showing Benchmark again, Test scaling and Set baseline controls.
A retained preview result with test conditions and baseline controls.

Measure frame times and spikes

Catch the frames
your averages hide.

Record available CPU and GPU frame timings in Play Mode. Inspect slow frames and timing history to understand stability against your target budget.

P50The typical frame
P95The slower tail
P99The rare spikes
  • Review timing history and spike markers.
  • Inspect selected moments in the recording.
  • Check frame-time headroom against your budget.

Timing availability depends on the platform and graphics backend. These are frame-level timings, not GPU time attributed to scene objects.

MEASURE / FRAME-TIME HISTORYGPUSight / Unity Editor
A recorded GPUSight Measure view with frame-time percentiles and spike history.
A real recorded timing view. Values retain the context of that recording.

Set performance targets

Define what “within budget” means.

Use built-in or custom profiles to set frame-time and workload targets. Profiles define your goals; they do not emulate hardware.

GPUSight Profiles view listing built-in target profiles and the selected profile settings.

Save, compare and share

Keep the before.
Understand the after.

Save snapshots, compare compatible evidence, and share JSON reports. Incompatible or unavailable results stay visible in the comparison.

GPUSight snapshot comparison with metric deltas and explicit unavailable-evidence messages in English.

One connected workflow

Find it. Understand it.
Check the change.

For solo developers, technical artists and game teams. Start with a first performance pass and follow the same evidence into a deeper investigation.

Find a place to start.

Use scene maps and grouped contributors to choose the object or material worth investigating.

Explore the maps ↗

Test your hypothesis.

Inspect object evidence, research the shader and benchmark a candidate material configuration.

Explore shader research ↗

Check what changed.

Record frame timings in the real scene, compare compatible snapshots and share your findings.

Explore measurement ↗

Know what the results mean

Good decisions need
clear evidence.

Scene facts
Geometry, LOD configuration, material settings and texture bindings give the finding context.
Estimates
Workload rankings, pressure indicators and source heuristics guide your investigation.
Measurements
Preview benchmarks and frame timings keep their conditions and availability visible.

Follow a visual guide

A screenshot at every step.

Go from your first scene investigation to material research, measurements and comparisons with the illustrated English user guide.

43 pages34 guided steps40 illustrations
Open the PDF guide English · GPUSight 1.0.0 · approximately 10 MB

A few useful details

Before you
dive in.

Requirements, measurement boundaries and where GPUSight fits in your workflow.

Talk to us on Discord
What does GPUSight help me optimize?

GPUSight helps you investigate rendering workloads: geometry and LODs, coverage and overdraw, materials, shader source, textures, and frame-time stability. Use its maps and breakdowns to choose what to investigate, then compare evidence after an intentional change. It does not automatically modify or optimize your assets.

Does it measure GPU time for every scene object?

No. Workload rankings, geometry shares and shader-source complexity guide investigation; they are not exact per-renderer GPU milliseconds. Material benchmarks measure a controlled preview workload. Play Mode timings are frame-level and are not attributed to individual objects. GPU pressure is not a GPU utilization percentage.

Which Unity versions and render pipelines are supported?

The current release workflow has been checked with Unity 2022.3.62f3 + URP 14.0.12 and Unity 6000.0.62f1 + URP 17.0.4. Scene View analysis includes Built-in, URP and HDRP paths, but individual features have separate requirements. These checks do not certify every Unity, pipeline or graphics API combination.

What are the requirements for GPU measurements?

Material preview benchmarks require active URP, a supported material and an available GPU timing backend. Native texture-allocation inventory currently requires Metal. Play Mode CPU and GPU timing availability depends on the platform, graphics API and timing source. The guide describes feature-specific requirements.

Do profiles emulate mobile or console hardware?

No. Built-in and custom profiles define performance targets and budgets. Selecting a profile does not emulate hardware or move the benchmark to another device. Controlled preview tests run on the actual host device and graphics backend.

Does GPUSight replace Unity Profiler or RenderDoc?

GPUSight complements Unity Profiler, Frame Debugger and RenderDoc. It connects scene investigation, object evidence, shader research and comparisons. New RenderDoc deep capture and remote Player Session workflows are not available in this release.

Can I share results with my team?

Yes. Export and import snapshot JSON to share captured evidence and its context, or export material benchmark reports. Snapshots preserve available results and identify incompatible comparisons; they do not package your Unity project or create a new measurement.

GPUSight for Unity

Make your next optimization
an informed one.

From a scene-wide pattern to a specific material.
From a hypothesis to evidence you can compare.

GPUSight screenshot