Expose the frame-time distribution
The first signal should reveal spikes, tails, and pacing instead of reducing the session to an average that hides stutter.
Performance diagnosis
A PC performance debugging stack from symptom to evidence
Separate frame pacing, CPU, GPU, system tracing, and graphics calls so optimization is driven by reproducible evidence rather than average FPS.
01
The problem to solve
A game that feels uneven may be limited by pacing, CPU spikes, GPU pressure, resource uploads, or operating-system scheduling, and average FPS cannot distinguish them. This collection selects tools around reproducible capture so the team classifies the bottleneck before testing the smallest useful change.
Selection criteria
Why these tools belong together
02
Narrow the ownership boundary
CPU, GPU, driver, system events, and graphics calls should be investigated in layers before a large subsystem is rewritten.
Make captures comparable
Record the baseline, scene, settings, hardware, and build so a before-and-after result can be attributed to the change.
Recommended workflow
Prove the path before scaling production
03
Freeze a reproducible test segment
Use the same save, camera, input sequence, resolution, and quality settings, then warm up and record hardware plus build identifiers.
Begin with low-overhead observation
Decide whether the symptom resembles sustained GPU pressure, a CPU limit, or an intermittent spike before taking a deeper capture.
Sample the responsible domain
Use timelines and zones for CPU work, frame capture for GPU events, and scheduler traces for system behavior instead of collecting everything.
Test one hypothesis at a time
Record the change, expected metric, and possible side effects, then rerun the same segment and inspect averages, percentiles, and worst frames.
Selected resources
Build the smallest useful combination from these six
The six tools progress from lightweight overlays through CPU and system timelines to GPU frame capture. Choose the next layer from the symptom already observed instead of enabling every profiler together.
Open sourceFreeCommercial
Tool
RenderDoc
Open-source frame debugger spanning major graphics APIs
RenderDoc captures a rendered frame and exposes draw calls, pipeline state, resources, meshes, shaders, and outputs across major desktop APIs and Android.
Open sourceFreeCommercial
Tool
Tracy Profiler
Low-overhead open-source CPU, GPU, and memory profiler for real-time software
Tracy combines client instrumentation with a standalone viewer for frames, thread zones, locks, call stacks, allocations, and GPU timing in native code and custom engines.
Open sourceFreeCommercial
Tool
Perfetto
Analyze cross-process system performance on one timeline
Google's open tracing platform provides Android/Chrome capture, a native SDK, SQL queries, and a browser timeline.
Open sourceFreeCommercial
Tool
apitrace
Record and replay OpenGL and Direct3D calls
An open graphics-API tracer capturing call streams, inspecting state, replaying frames, and debugging pixels and shaders.
Open sourceFreeCommercial
Tool
MangoHud
Overlay frametime and hardware performance metrics on Linux games
An open Vulkan and OpenGL performance overlay showing FPS, frametime, CPU or GPU load, temperatures, memory, power, and other metrics with logging, presets, and local plotting.
SubscriptionTerms apply
Tool
Superluminal Performance
Locate native game CPU hotspots with a low-overhead sampling profiler
A commercial profiler for games and real-time applications that samples multithreaded native code on Windows, Linux, and consoles with call trees, timelines, locks, context switches, and custom events.