Population performance reference results

These measurements are the first published Mass Forge simulation baseline. Each report was produced by the packaged Run-MassForgeBenchmarks.ps1 workflow with five measured iterations after one warm-up, the default 10k/100k/1M populations, NullRHI, and exact per-entity result validation.

Reference machine and conditions

• Windows 11, OS build reported by Unreal as 10.0.26200.1.256.64bit.
• Intel Core i9-14900K, 24 physical cores and 32 logical cores.
• 102,747,594,752 bytes of physical memory reported by Unreal.
• Development Editor commandlet in a disposable host project, immediately after strict Editor, Development Game, and Shipping Game package builds.
• Actorless Mass simulation only. No rendering, VAT, navigation, representation, project AI, or gameplay outside the measured Mass Forge path.

The implementation under measurement is the benchmark/high-volume-effects source shipped with these reports. The raw CSV metadata records exact engine versions and timestamps. Future comparisons must also record the plugin commit because timings can change independently of the public plugin version.

Published results

Times are milliseconds. Payload and observed physical delta are MiB. Queued mean/worst is mailbox submission plus built-in processor execution.

Engine	Entities	Payload	Observed delta	Create	Direct mean	Direct worst	Queued mean	Queued worst
UE 5.6.1	10,000	3.82	6.05	1.551	0.322	0.331	0.374	0.379
UE 5.6.1	100,000	38.15	41.30	11.710	3.312	3.486	3.887	3.933
UE 5.6.1	1,000,000	381.47	402.65	169.607	34.464	37.072	43.464	47.039
UE 5.7.1	10,000	3.82	6.11	1.527	0.420	0.852	0.362	0.374
UE 5.7.1	100,000	38.15	41.41	10.856	3.145	3.147	3.603	3.641
UE 5.7.1	1,000,000	381.47	401.57	116.537	33.935	34.953	39.011	39.845
UE 5.8.2	10,000	3.82	5.93	1.462	0.313	0.318	0.368	0.377
UE 5.8.2	100,000	38.15	42.96	11.099	3.157	3.196	3.689	3.704
UE 5.8.2	1,000,000	381.47	402.52	115.527	32.698	34.584	38.166	38.503

Every row passed. The deterministic payload is about 400 bytes per entity for Vital, Combat, Support, and the high-volume mailbox in this benchmark archetype, plus one 2,388-byte shared compiled-effect table. The observed delta includes Mass chunk and allocator overhead and is not a replacement for the deterministic fragment-capacity report.

At one million entities, the serial compiled pass costs roughly 33–34 ms mean on this machine and therefore does not fit a 60 Hz frame if every entity receives an effect every frame. That is a useful capacity boundary, not a failure: real projects should distribute effect work, use Mass LOD/significance, and measure their actual archetypes and presentation workload. The 100k path remained below 4 ms mean and worst in all three engine runs.

The UE 5.7 10k direct worst sample is an isolated outlier relative to its mean and the other engines. It is retained rather than edited out; longer runs are appropriate when setting a production budget.

Raw reports

• UE 5.6.1 CSV
• UE 5.7.1 CSV
• UE 5.8.2 CSV

See the benchmark guide for reproduction, interpretation, and the explicit rendering/worker-thread boundary.