Hello Creators,
We’ve recently made many improvements to the MicroProfiler. The latest update introduces detailed memory profiling, flamegraphs, and dump comparisons.
Key features include:
- X-Ray mode for visualizing memory allocations and identifying areas with excessive memory operation intensity.
- Summary FlameGraphs for an aggregated view of CPU and memory usage, aiding in understanding overall performance.
- Diff FlameGraphs for easy comparison of two dumps, highlighting performance changes.
Performance and usability improvements:
- Five times more compact dumps.
- Improved browser scaling support and frame time bottleneck information.
- Added Re-capture and Save To File buttons, along with a Ctrl+F hotkey to find a timer by name.
These changes make MicroProfiler a more powerful and easier-to-use tool for performance analysis and optimization.
- For a more detailed guide on using the new features, see the full article.
- If you’re just getting started with the MicroProfiler, check out the documentation for an overview.
Watch the video below for an overview of our updates:
If you have any ideas for further additions, please let us know in the comments. Thank you for using MicroProfiler and caring about performance!
Happy building!
ZenMa1n
Appendix
X-Ray Mode for Memory Profiling
The detailed picture of memory allocations/deallocations is now accessible for individual scopes. We aimed to keep the interface simple, so allocation intensity is displayed as an overlay using shades of gray, hence the name X-Ray mode.
The two gray bars at the top allow you to spot areas with excessive memory allocation, even if you were profiling CPU rather than memory. If you notice such areas in a frame, switch to X-Ray mode by pressing the X key on your keyboard or by selecting X-Ray → Main View from the top menu. CPU scopes will turn grayscale and you’ll see which ones are experiencing the most allocations, either by number or total size (yes, those will appear brighter than the others).
Summary FlameGraph for CPU and Memory
In detailed mode, it’s not always immediately clear what the overall CPU or memory usage looks like, especially in cases where a particular task consumes few resources but is called very frequently, subtly affecting the entire frame. For these situations, we created a mode that aggregates all individual call stacks into one large graph – this is the Summary FlameGraph.
You can find them in the top menu under Export → CPU Flamegraph and Memory Flamegraph. When you click on one of them, a new browser tab will open (or prompt you to download a file). The CPU and memory Flamegraphs look similar. When you hover over a scope, detailed information about it will be displayed at the bottom, including total CPU time, the number of memory allocations, or summarized allocation size depending on the selected mode. Click on a scope to zoom in.
It’s important to note that in these Flamegraphs, the statistics are shown only for the frames included in the dump, and not for the entire duration since the start of the experience.
Diff FlameGraphs
When we make changes to the experience, it’s nice to understand whether they lead to performance improvements or degradations without squinting at numbers, right? For this, we’ve introduced Diff FlameGraphs.
This is where working with MicroProfiler in the browser is a more powerful tool than working in the Roblox Player or Studio window. Since we save dumps to disk, we can track progress in improving performance over time (over weeks, months, from version to version, etc.).
Open one dump and drag the second dump (HTML file) directly into the browser window — a popup will appear where you can drop the second dump (you can also access this window from the top menu by clicking Export → Diff / Combine).
In the Left section (marked in Green), the name of the currently opened dump will appear automatically, while you drop the second dump in the Right section (marked in Blue). Click Combine & Compare, and in a new tab (or as a downloaded file), you’ll get a comparison like this.
The color of the scopes here depends on which dump (Left/Green or Right/Blue) consumes more CPU/Memory resources. The brighter the color, the greater the difference.
The displayed values are averaged per frame. This is because the dumps being compared can contain a different number of frames (for example, 32 and 128), and we still want to compare them, so we calculate the values for one averaged frame from both sides before comparing.
Five Times More Compact Dumps
We’ve revamped the data representation format (though it’s still an .html page), and now when taking a 512-frame dump, the size on disk may now be around 17MB, compared to 90MB previously.
Re-capture and Save To File Buttons
When opening a capture via HTTP (directly from a mobile device), you will see two new buttons in the top menu. The first button, Re-capture, allows you to take a new capture without issues related to the browser cache (pressing F5 may sometimes show the old page from the cache instead of the new dump). The second button, Save to File, lets you save the open capture as an HTML file while preserving its full name, including the timestamp, without worrying about the browser distorting the original page upon saving.
Improved Browser Scaling Support
Previously, the MicroProfiler interface would break when zooming the page. Now, it remains proportional and usable. This improvement is especially beneficial when demonstrating dumps to others, such as during video calls or presentations.
Frame Time Bottleneck Cause
Some time ago, we introduced color-coded highlighting for frames in the top part of the Profiler to indicate their performance characteristics. The purpose is to distinguish frames categorized as CPU-heavy (orange), GPU-heavy (red), or Render-heavy (blue).
When you hover over a frame, a tooltip is displayed. We have added detailed timing information to these tooltips to clarify how a frame is categorized. The value that most significantly contributes to the frame’s classification is highlighted in red.
Ctrl+F (Cmd+F)
We now have the ability to search for scopes by name using a more familiar hotkey. When you press Ctrl+F on Windows (or Cmd+F on Mac), a search box will appear in the lower-left corner. Type the name of the scope there, press Enter, and you’ll be taken to the instance of the scope that takes up the most time in that dump.
Reference Frame Time Adjustment
When profiling a game on a slow device, the vertical bars representing frame times at the top of the page can sometimes shoot up, making it unclear which frame is faster and which is slower (since technically they’re all slow). Now, if you hover over these bars and scroll your mouse wheel, you can adjust the upper limit of frame time so that all values fit on the screen without being cut off at the top. This is a faster method than navigating through the top menu and selecting Reference → 50 ms (100 ms, etc.).
Group Highlighting
A new option has been added to the top menu: Highlight → [group_names]. When you select a group (like Physics or Script), the scopes that belong to that group will continue to display as usual, while all other scopes will turn gray, reducing distractions. Selecting Highlight → None will return everything to its original state.
Scope Labels Improvements/Fixes
In the past, large captures (256 or 512 frames) with many scripts would occasionally overflow the internal buffer for labels containing script names. This would result in earlier saved frames displaying scope names as “Script_null”, while later frames appeared correctly. We have optimized how this buffer works and this should significantly reduce instances of overflow and ensure all script names are displayed correctly.
