CS2: Steam Overlay Off vs On — build 1791918634

With Steam build 1791918634, overlay ON recorded 1.00% higher median average FPS, 7.12% higher 1% low and 0.61% lower p99 frame time. Repeats disagreed and ranges overlapped. Disabling overlay did not demonstrate a 5–10% FPS gain; the observed differences do not establish a causal benefit in either direction.

Input latency: not measured. FPS and these source-specific timing metrics do not establish click-to-photon latency or guarantee smoother gameplay.

Test setup & definitions

Steam client build

1791918634

Steam file version

11.12.30.07

Overlay DLL version

11.12.30.07

Version evidence

Reconstructed after capture from the same uninterrupted Steam session: startup/update log, installed executable and overlay DLL. The newer downloaded manifest remained pending, not the tested client.

CPU

AMD Ryzen 5 7600X3D

GPU

NVIDIA GeForce RTX 4060

Driver

617.42 (Windows 32.0.16.1742)

OS

Windows 11 Pro 10.0.26300

CS2 build

26087496

Scene

FPSHeaven Ancient · workshop 3472126051

Display

1440 × 1080 · 144 Hz

Fixed controls

Original NVIDIA profile, including ReBAR and Reflex, held unchanged; Steam performance/FPS monitor OFF

Capture

PresentMon 2.6.0 · fixed60-offset10-v1

Steam global overlay was ON for both blocks. Only the CS2-specific Steam Overlay preference changed (explicit OFF/ON), checked through Steam UI and saved-preference readback. CS2 relaunched for every test; its overlay panel remained closed during capture. No Shift+Tab behavior pilot was used.

OFF: one warmup then four measured runs. ON: one warmup then four measured runs. This block order was requested for the direct test; it is not ABBA or randomized. All ten recordings are retained, including the slower first OFF measurement. Warmups do not enter summaries.

PresentMon covers exactly 60 seconds, starting 10 seconds after the observed VProf start. Average FPS = 1,000 / mean consecutive CPU frame-start interval in ms; 1% low = 1,000 / mean of the slowest ceil(1%) intervals; p99 uses the nearest-rank 99th percentile.

The existing descriptive rule requires at least a 2% median difference, four same-direction repetition-rank differences and non-overlapping ranges. No metric passed. Rank matching across these blocks is not a contemporaneous paired design and cannot remove time/order drift.

This report identifies the tested Steam build. It compares overlay OFF and ON within one client version; it does not compare different Steam versions or predict results after an update.

Data quality & downloads

All 10 workload recordings completed: two warmups and eight measurements. The collector completed after its documented stop marker was issued. Independent offline analysis replayed recorded timing boundaries, foreground samples, startup evidence, Steam conditions and fixed NVIDIA snapshots; CSV hashes were checked before and after analysis.

The runner stalled after capture; its precise shutdown cause remains unresolved. The owned runner was stopped and raw data preserved. Offline analysis used the existing PresentMon calculations; no benchmark was rerun to replace results.

NVIDIA, Windows and all ten CFG files matched original state after recovery. Steam restoration remained incomplete when Computer Use was stopped: the global overlay preference was still ON instead of its original OFF. This is disclosed separately from the completed capture validation.

Steam version was reconstructed after the run, not continuously fingerprinted during every capture. The running Steam session began before measurement; updater logs identify installed build 1791918634 and a newer download awaiting restart.

• One PC, one synthetic Ancient scene and four measurements per state; not a prediction of match FPS.
• OFF was tested before ON. Cache, temperature and background activity may contribute to differences.
• The first measured OFF run was slower (275.96 FPS; 67.09 FPS 1% low) and remains included. Its cause was not established.
• Ranges overlap and rank-matched differences disagree; a roughly 1% average-FPS difference is not evidence that enabling overlay improves performance.
• The 7.12% median 1% low difference is also inconsistent across repeats and is not proof of smoother play.
• Input latency was not measured. VProf P1 retains its source definition and is not the PresentMon 1% low.
• Results apply to the recorded Steam and CS2 builds; future updates require new measurements.

Downloads contain allowlisted derived measurements, including warmups. Original captures, account identifiers, local paths, device identifiers and configuration contents remain private.