Video summary

Which Frame Limiter Has the Lowest Latency? RTSS vs Special K vs Radeon Chill

Main summary

Key takeaways

Technology

Summary of the video (tech concepts + findings)

Goal: compare frame limiters and real input latency

The video investigates how different frame limit / VRR synchronization methods affect not only 1% low/frame-time smoothness, but also real end-to-end input latency.

A central claim being challenged is that extremely good frame-time graphs might imply “insane” input lag—potentially due to buffering.

How the “perfect” frame-time graph is achieved (RivaTuner guide)

The speaker uses RivaTuner Statistics Server (RTSS) and emphasizes that the key is the Frame Limiter.

Core method:

  • Set the RTSS frame limit below your monitor’s max refresh rate and below your typical/average FPS.
    • Trade-off: you sacrifice some headroom to smooth frame delivery.

Graph reset / recording workflow:

  • When using MSI Afterburner/benchmark recording, use a hotkey (example given: Shift+F4) to reset/record graphs consistently.

Limiters being tested

RTSS (4 limiters)

  • Async (default)
  • Front Edge Sync
  • Back Edge Sync
  • Nvidia Reflex (speaker states it works with AMD too)

Special K (3 limiters)

  • Low latency (optimized for VRR)
  • Normal (higher latency than low latency)
  • Latent Sync (optimized for no V-Sync and no VRR)

Radeon Chill

Unique behavior: two different limits depending on input state:

  • One limit when the mouse/camera is moving
  • Another limit when it stops

The speaker couldn’t achieve perfectly consistent (“perfect”) results, though a smooth frame-time graph was still feasible.

Important warning about “numbers”

The video repeatedly cautions that latency/frequency metrics can be misleading because:

  • Latency varies with loading, scene changes, menus, etc.
  • Different tools (e.g., Special K) may compute 1% low differently.

How latency was measured (hardware + methodology)

Instead of slow high-speed-camera methods, the speaker used an external latency measurement device from OSRTT (osrtt.com).

Testing method:

  • The device visually triggers in-game events (e.g., simulated button press / small mouse movement) and measures end-to-end latency.

High-volume testing:

  • Runs 50 tests every ~0.5 seconds
  • Total: 150 tests in the main run

Display context:

  • A 120 Hz FreeSync/VRR environment
  • The speaker explains that the timing window (roughly 8 ms per refresh) makes “perfect alignment” difficult.

Key latency findings (RTSS/VRR/sync behavior)

1) FPS vs refresh rate: refresh matters

They tested scenarios including:

  • 30 FPS on a 30 Hz scenario (worst)
  • Uncapped / ~300 FPS on 120 Hz (best latency, but tearing risk)

They also note that “minimum/maximum/average” differ because:

  • A frame may not perfectly land within the refresh window, affecting latency readings and potentially causing tearing.

2) The “best latency” scenario was often unplayable due to screen tearing

Example:

  • ~300 FPS @ 120 Hz produced the best measured latency but caused noticeable tearing.

Solution approach:

  • Use VRR + a lower frame cap (e.g., cap below max like 116 or lower) to prevent tearing while keeping latency low.
  • The speaker highlights RTSS Async as a key component in the “no tearing + low latency” setup.

3) V-Sync behavior is not uniformly terrible

Major takeaway: V-Sync is massively latency-heavy only at/near maximum refresh, but much less harmful below the max.

They argue against the common belief that V-Sync/triple buffering is always “insanely heavy.”

Observations:

  • V-Sync at max FPS (e.g., locking behavior around 220 FPS) produces heavy latency (described as “triple” feeling)
  • V-Sync below max refresh behaves much closer to “free” latency-wise

4) VRR-only vs adding additional sync options

They compare RTSS cap + VRR enabled (example: 116 FPS to avoid tearing).

General trend:

  • Async vs Front Edge / Back Edge: adding certain RTSS edge sync methods increased latency relative to Async
  • Nvidia Reflex was reported as one of the fastest RTSS options tested
  • Radeon Chill looked almost identical with V-Sync enabled vs disabled

Special K insights

Special K limiters show behavior similar to RTSS methods when used with VRR.

Key claims:

  • Special K “Normal” limiter was said to be as heavy as RTSS “Front Edge” (or very similar).
  • Special K “Latent Sync” (designed for no VRR/no V-Sync) showed very low minimum latency, and in this test it looked unusually good.

Linux note (bonus)

The speaker says the same latency test app doesn’t work on Linux, so they couldn’t replicate the exact hardware latency comparisons.

However, they believe Linux smoothness feels at least as good or better from experience.

High-level Linux setup tutorial:

  • Use Steam launch options with MangoHud
  • Mention overlay tools like Goverlay / Cas OS-related options
  • Frame limiting with parameters such as:
    • Example: offset -4
  • Choose limiting “method” (speaker prefers early because it stabilizes 1% lows and smooths the graph)
  • Enable Adaptive Sync/VRR automatically (examples mentioned: KDE choosing “automatic”)

Distro notes:

  • On SteamOS/Bazzite, they mention a “console-like” performance tab to set FPS below max and enable VRR.

Main speakers/sources (end of video)

  • Main speaker: the channel host (not named in the subtitles)

Referenced external sources/channels:

  • Gamers Nexus (used as a conceptual reference for latency definition/testing rationale)
  • Battle(non)sense (previous/high-speed-camera/robot-arm latency measurement approaches)
  • Techless (mentioned regarding V-Sync latency behavior)
  • OSRTT (osrtt.com) (the latency measurement device)

Software referenced by name:

  • RTSS (RivaTuner Statistics Server)
  • Special K
  • Radeon Chill
  • MangoHud
  • Goverlay / Cas OS
  • MSI Afterburner (for hotkey/benchmark recording)

Original video