DEV Community

JuegoTech
JuegoTech

Posted on

Unreal Insights in the Sprint: A Short Performance Pass You Can Finish Before Standup

Most Unreal teams only open Unreal Insights when something is already on fire: a milestone build hitches, QA files a "feels choppy" bug, or a platform target slips. By then you're profiling under pressure, against a build that has drifted for weeks.

A better habit is a short, repeatable pass that one engineer can run on the latest build the afternoon before standup. It won't replace a deep optimization sprint, but it catches regressions while the change that caused them is still fresh in someone's head.

Here's the pass I'd recommend. It assumes UE5 and a development or test build on your target hardware, not the editor.

1. Capture the same route every time

Pick one fixed, repeatable route through the game: a 3-5 minute path that hits a dense area, a streaming transition, a combat or heavy-VFX moment, and a menu open/close. Write it down so anyone on the team can run it.

Launch with tracing enabled, for example:

YourGame.exe -trace=default,memory -statnamedevents
Enter fullscreen mode Exit fullscreen mode

-statnamedevents gives you more readable scopes in the timing view; drop it if the overhead skews your numbers on low-end hardware. If you'd rather start and stop manually, Trace.Start and Trace.Stop in the console work too.

Drop bookmarks at the start of each route segment (Trace.Bookmark SegmentName) so you can jump straight to them later instead of scrubbing.

2. Frame time first, not averages

Open the capture in Timing Insights and look at the Frames track before anything else.

  • Look for spikes, not the mean. A build that averages fine but drops a long frame every few seconds will feel worse than one that's steadily a bit slower.
  • Check whether the spike lives on the Game thread, Render thread or GPU. That single answer decides who picks up the ticket.
  • Compare against the previous capture of the same route. If you don't have one yet, today's capture becomes the baseline.

3. Hitches: find the first long frame and zoom in

For each hitch, select the frame and expand the timing tree. Common culprits worth checking every time:

  • Asset loading or streaming on the game thread (synchronous loads that slipped in with new content).
  • Shader compilation / PSO creation the first time a material or effect is seen. Note which effect triggered it; this is usually a precaching or warm-up task, not a code bug.
  • Garbage collection landing in the middle of gameplay.
  • Blueprint tick costs that crept up as designers added logic.

Write down the scope name and the frame number. That's enough for a ticket.

4. Memory: one glance at the trend

Open Memory Insights (from the memory channel) and look at the overall trend across the route.

  • Does memory return close to where it started after you leave the dense area, or does it keep climbing?
  • Which LLM tags grew the most compared to the last capture? Textures and meshes after a content drop are expected; an unexplained climb in a gameplay system is not.

You don't need to solve a leak in this pass. You need to notice it the day it appears.

5. File one ticket per finding, with the trace attached

Keep the output small and actionable:

  • Route segment and bookmark name
  • Frame number or time range
  • Thread (Game / Render / GPU) and the top scope
  • Link to the .utrace file and the commit or changelist range since the last capture

Then save the capture as the new baseline for that route.

Why this works in a sprint

The whole pass fits in well under an hour once the route is written down, and because the route and launch arguments are fixed, captures are comparable week to week. Regressions show up as diffs against last week instead of mysteries before a milestone.

If your team is short on engineers who are comfortable reading Insights traces, it's worth pairing the pass with someone who's done it on shipped titles. We do this kind of profiling work as part of our Unreal Engine development projects at Juego Studios, and it's worth running the same pass before a live event, alongside a short live event go-live checklist.

What does your team's profiling routine look like? I'd like to hear what other studios put on their fixed capture route.

Top comments (0)