Tracy Profiler

Overview

Tracy Profiler is a real-time tool for profiling games and other applications through remote telemetry, frame profiling and sampling at nanosecond resolution. It suits developers who need to inspect CPU and GPU work alongside system-level events, rather than only monitor frame rates. Its broad capture scope is a clear strength, but merged traces do not retain every captured data type.

Released under the 3-clause BSD license, Tracy pairs instrumentation with a profiler interface and supports local analysis of captured data through an AI assistant integration. It is a focused engineering tool, best suited to teams able to work directly with profiling integrations and trace files.

Key features

CPU and GPU coverage

Direct CPU profiling integrations cover C, C++, Lua, Python and Fortran, while the project points to third-party bindings for Rust, Zig, C#, OCaml and Odin, among other languages. GPU support spans OpenGL, Vulkan, Direct3D 11 and 12, Metal, OpenCL, CUDA and WebGPU. That range makes Tracy a practical option for projects with varied graphics stacks; teams using languages outside the direct integrations should account for the third-party nature of those bindings.

Trace detail and capture

Along with frame-time analysis, Tracy can capture memory allocations, locks and context switches, and associate screenshots with captured frames. This combination helps developers relate performance events to both system activity and visual output. Its remote profiling support and 0.14 capture daemon extend that workflow: the daemon discovers Tracy clients over UDP and saves each capture to its own file.

Merging and sampling

The 0.14 release also added a tool to combine multi-process traces. The merge retains zones, messages and plots, but drops GPU zones, memory events and call stacks. That is a meaningful constraint for investigations where those data are central: keep the separate captures when their full detail matters. With sampling enabled, Tracy can heuristically reconstruct zone call stacks, which adds context but should not be treated as a substitute for data the merge discards.

Local assistant

An assistant integration uses llama.cpp models to analyze captured Tracy data and can run locally after setup. Keeping that analysis on the user's system is useful for developers who want a private workflow, though it requires local configuration rather than being an automatic, hosted feature.

Pricing

Tracy Profiler is free: the plan costs 0.00 USD per free under the 3-clause BSD license. There is no paid tier or trial to weigh against a free allowance, so the central trade-off is technical fit rather than plan limits. The license permits use, copying, modification, publication and distribution.

Platforms

Supported platforms include Windows, Linux, Android, FreeBSD, WSL, macOS, iOS and QNX. Prebuilt profiler binaries are provided for Windows, macOS and Linux; GitHub releases also include compiled Windows x64 binaries and the user documentation as a PDF. Users of other supported platforms should distinguish platform support from the narrower availability of prebuilt profiler binaries.

Who it's for

Tracy is a strong fit for game and application developers who need detailed CPU, GPU and system telemetry, remote capture, or profiling across multiple processes. Its wide API coverage and free BSD licensing suit technically equipped teams that want to instrument their own software. It is a weaker fit for readers seeking a simple frame-rate monitor, or for workflows that depend on preserving GPU zones, memory events and call stacks in merged traces.

Pros and cons

• Pro: Broad telemetry in one profiler. CPU, GPU, memory, locks, context switches and frame-associated screenshots support investigations beyond frame timing alone.
• Pro: Wide graphics API and language reach. Direct integrations and third-party bindings cover many common development stacks.
• Pro: Free under a permissive license. The 3-clause BSD license allows modification and redistribution.
• Con: Merging loses important trace data. GPU zones, memory events and call stacks do not survive multi-process trace merging.
• Con: The local assistant needs setup. Its privacy advantage comes with the responsibility of configuring it on the user's system.
• Con: Prebuilt profiler binaries cover fewer platforms than the support list. Windows, macOS and Linux have prebuilt binaries; other supported platforms may require a different workflow.

Alternatives

For a broader shortlist, see Game Profiling Software.

• GameBench is worth considering for readers who want a paid product with a 30-minute Pro trial and unlimited sessions for 28 days.
• Arm Performance Studio is another free choice, particularly for profiling Arm CPU and GPU applications.
• RAD Telemetry is an alternative with custom telemetry licensing.
• Optick is a free profiler under the MIT License.
• RenderDoc is a free MIT-licensed option for readers comparing tools in this category.
• AMD Radeon GPU Profiler is a free alternative for users with a supported AMD Radeon GPU and operating system.
• MicroProfile is an embeddable C++ profiler to consider when that format suits the project.
• Unity offers a free tier with pay-as-you-go usage beyond included limits.

Verdict

Choose Tracy if you need detailed, remotely captured CPU, GPU and system traces across a broad set of graphics APIs, and can work within its instrumentation-led approach. Its free BSD license and extensive telemetry make it compelling for serious game and application profiling. Look elsewhere when merged captures must preserve GPU, memory and call-stack data, or when you need a less technical profiling workflow.