PerfCop - Documentation
Help and FAQs for PerfCop!
Frequently Asked Questions
What Is PerfCop?
PerfCop is a service that helps people to analyze performance problems and bottlenecks with their Unreal Engine projects.
How Does It Work?
Basically, you use the Unreal Insights performance capture system to record a trace (utrace) file.
Once you have that trace file, you can upload it to PerfCop and get some useful intelligence into what your performance bottlenecks or problems are and how you can go about fixing them.
For more information, please refer to the Getting Started page.
What Sort Of Performance Trace Channels Are Supported?
At the present time, we support bookmark, log, frame, cpu and gpu trace channels. We don’t, as yet, support memory or stack channels but this is coming soon.
For a full list of the available trace channels please refer to the Unreal Documentation.
What Are Credits?
In an ideal world, everything would be free but then it wouldn’t be worth having, would it?
Credits are how PerfCop provides the service. As you perform uploads, generate reports, participate in Sherlock chats, etc, these actions will consume credits.
As part of your monthly subscription, you will be granted credits for you to use accordingly to the level of Plan purchased by your organization.
During this pilot phase of PerfCop, we’re evaluating just how Credits are going to work, how they’re consumed and what levels are necessary to provide the level of service needed to our customers.
Right now, you shouldn’t really ever run out of Credits but if you do, please contact Support and we’ll be happy to top up your credits for you!
Who’s Sherlock?
Sherlock is PerfCop’s chat bot, of course! If you have a question or query about an aspect of a report, use Sherlock to help provide or clarify an answer!
Give it a try - it’s soooo useful!
On The Projects Dashboard How Is The Trace Size Value Calculated?
When working with uploads, the Trace Size is based on the size of the data (utrace) we ingest along with associated meta data and other storage data.
This means that you could be looking at a 2x or 3x size increase above the original utrace file size (e.g. 200MB utrace is 400-600MB of Trace Size data).
Additionally, reports and any Sherlock data can increase this but not substantially.
If you’re concerned about Trace Sizes, please contact Support with any queries you might have.
Bugs? What Do We Do If We Find A Bug?
Well, firstly, don’t tell anyone otherwise they’ll all want one!
However, on a more serious note, PerfCop is a Closed Alpha Pilot Program service and, as such, will have its fair share of issues, gotchas and not-so-clear elements.
We actively encourage feedback so we can make PerfCop the best that it can be so head on over to our Support page for more details!
How Does Privacy Work In PerfCop?
PerfCop is designed to analyze Unreal Engine performance while keeping your proprietary project data private and inaccessible to us. We achieve this using a layered privacy model:
-
Trace files may be stored, but we cannot view them.
When you upload a utrace file, it is stored inside a secure, isolated environment. PerfCop engineers and support staff have no ability to open, inspect, or browse the contents of these trace files - even though they may be retained temporarily for processing. -
Scope names are always anonymized inside our systems.
Unreal Engine’s profiling “scopes” are collected from the public engine source code and transformed using a one-way, irreversible hash function. When we ingest your trace, we apply the same hashing process. This lets us analyze usage patterns, but makes it impossible for us to read or reconstruct your actual scope names — whether they come from the engine or from your project. - So how do reports display real scope names?
PerfCop itself never sees your readable scope names, but the report generator needs them to produce human-friendly output. To solve this, we securely send your unfiltered project data to Google’s enterprise AI platform, which:
- Receives the data through a private and encrypted channel
- Generates the natural-language portions of the report
- Returns the human-readable output to you
- Is contractually prohibited from training on, storing, or using that data for any purpose beyond the single request
PerfCop never receives the readable scope names - only Google’s final text output, which contains the explanations and recommendations.
- Internally, all analytics stay anonymized.
For model training and trend analysis, PerfCop only uses:
- Hashed scope identifiers
- Usage frequencies
- Correlation patterns
- Aggregated statistical data
We never use raw project identifiers, readable scope names, asset references, or code-related strings.