async-profiler vs Scalene

async-profiler is a free, self-hosted sampling profiler for Java designed to keep overhead low and avoid safepoint bias. It works with OpenJDK and other runtimes based on the HotSpot JVM, and stack traces can include non-Java threads as well as native and kernel frames. It profiles CPU time, Java-heap and native-memory allocations, memory leaks, contended locks, and hardware or software performance counters. Documented modes include CPU, allocation, wall-clock, Java-method, and multiple-events profiling. Results can be saved as collapsed stacks, flame graphs, HTML tree views, or text; an interactive Flame Graph can be viewed in a browser. Officially maintained builds are available for Linux and macOS on x64 and arm64. Developers can use the Java API as a Maven dependency, and IntelliJ IDEA includes async-profiler in its Java Profiler settings. It can profile Java processes in Docker or LXC from inside a container or from the host. Building it requires make, GCC 7.5.0 or later or Clang 7.0.0 or later, static libstdc++, and JDK 11 or later.

Scalene is a free, open-source profiler for Python programs that reports CPU, GPU and memory use and offers AI-powered optimization proposals. Its line- and function-level reports separate Python time from native-code and system time. Memory analysis shows per-line use, distinguishes Python from native allocations and flags lines with likely memory leaks. GPU profiling reports time on NVIDIA-based systems. Async profiling attributes wall-clock await time to the suspended coroutine line and reports mean and peak concurrent waiters. Its graphical interface can display stitched Python and native call stacks, memory flame charts and a timeline. Scalene is usable from the command line, a web-based GUI, a Visual Studio Code extension or Jupyter notebooks. The web interface processes profiles locally and works offline. For optimization suggestions, listed AI providers include Amazon Bedrock, Microsoft Azure, OpenAI and local models through Ollama. The project reports typical profiling overhead of no more than 10–20%, often less. Scalene is available for free under the Apache-2.0 license, with installations listed for macOS, Linux, Windows and Windows Subsystem for Linux 2.