Gear GeeksGaming

Reflex, Anti-Lag and the render queue

This is the highest-value setting in competitive PC gaming and it is a checkbox. Understanding why requires understanding the render queue, which is the one part of the latency chain that can be dozens of milliseconds long without anything being broken.

LatencyUpdated 7 min read

Why a queue exists at all

A GPU is expensive and should never sit idle. To keep it fed, the CPU works ahead: it simulates a frame, hands the draw commands to the driver, and immediately starts on the next one. If the GPU is slower than the CPU, those prepared frames stack up in a queue.

The queue does exactly what it was designed to do. It smooths over variation in CPU work, and it keeps average framerate high. The cost is that every frame sitting in the queue was simulated using input from the moment it entered the queue, which may be several frames ago.

queue latency = queue depth x frame time 3 frames at 60 fps = 50.0 ms 3 frames at 144 fps = 20.8 ms 1 frame at 240 fps = 4.2 ms

Fifty milliseconds is more than the entire rest of the local latency chain put together on a well-configured setup. That is the size of the problem.

What Reflex and Anti-Lag change

Both technologies do essentially the same thing under different names: they pace the CPU so that it starts simulating a frame just in time for the GPU to render it, rather than as early as possible.

The queue with and without frame pacing, GPU-bound at 144 fps.
ConfigurationTypical queue depthQueue latencyFramerate impact
Uncapped, no pacing, GPU-bound2 to 3 frames14 to 21 msHighest average
Framerate capped below GPU limitUnder 1 frameUnder 7 msCapped by definition
Reflex or Anti-Lag onAbout 0.5 framesAbout 3.5 msSmall reduction when GPU-bound
BothMinimalMinimalBest latency outcome

Note that these technologies do nothing to the rendering itself. They do not make frames faster, they make the frame that is being rendered newer. That is why the framerate cost is small and the latency benefit is large.

When it does nothing

Both are frequently reported as "not doing anything", and usually the report is accurate for a specific reason.

  • You are CPU-bound. The GPU is already finishing frames faster than the CPU produces them, so no queue exists. Nothing to remove. This is the common case in competitive titles at low settings.
  • You already cap your framerate well below your GPU's capability. The cap has already prevented the queue from forming.
  • Your framerate is limited by vsync at the refresh ceiling. Different mechanism, different fix: cap below your refresh rate.
  • The game implements it partially. Reflex requires engine integration; a title that only exposes the driver-level fallback gets less benefit.

In all of those cases enabling it still costs nothing, so leave it on. Just do not conclude the technology is fake because it did not help a configuration that had no queue.

The framerate cap, which is free everywhere

If your game or GPU does not support these technologies, a framerate cap achieves most of the same effect. The mechanism is identical: keep the GPU from being fully saturated, so frames never accumulate.

Choosing a cap.
SituationCap atWhy
Using variable refresh3 to 5 fps below refresh rateStays inside the VRR window, avoiding the vsync ceiling penalty
No VRR, framerate far above refreshNo cap, or well above refreshFresher frames, at the cost of tearing
GPU-bound, no VRR5 to 10% below your sustained framerateLeaves GPU headroom so the queue never fills

Set the cap in game where possible. In-game limiters cap before the frame enters the pipeline; driver-level limiters act later and are marginally worse, though still far better than nothing.

The configuration we recommend

  1. Reflex or Anti-Lag: on, in every game that offers it.
  2. Framerate cap: 3 to 5 below refresh rate if using VRR, otherwise below your sustained GPU limit.
  3. VRR: on, in both the monitor menu and the driver.
  4. Driver-level vsync: on, to handle the ceiling case gracefully. In-game vsync: off.

That combination is tear-free, keeps the queue empty, and lands within a millisecond or two of the theoretical best for your hardware.

Related reading

Frequently asked questions

What does Nvidia Reflex do?

It keeps the CPU from running ahead of the GPU. Instead of queueing frames the GPU is not ready for, the CPU is paced so it begins simulating a frame only when the GPU is nearly ready to render it. That means the frame being drawn was simulated with the most recent input available, rather than input from several frames ago.

Does Reflex reduce framerate?

Slightly, sometimes, when heavily GPU-bound, because pacing the CPU removes some of the buffer that smoothed over variation. The trade is almost always worth it: a few percent of framerate for a large reduction in latency.

Should I use Reflex if I am CPU-bound?

It does much less. The queue only fills when the GPU cannot keep up with the CPU. If your CPU is the bottleneck, the GPU is already waiting and there is no queue to remove. Enabling it costs nothing, so leave it on, but do not expect the same improvement.

Is a framerate cap as good as Reflex?

Close, for the same reason: capping below your GPU's maximum means the GPU always has spare capacity, so frames never accumulate. Reflex is more precise, adapts automatically as scene complexity varies, and does not require you to find the right cap. Use both.

What is Reflex Low Latency Boost?

An additional mode that keeps GPU clocks high even when the GPU is not fully loaded, so it responds faster to a sudden burst of work. It reduces latency slightly further at a cost in power and heat. Worth enabling on a desktop, less obviously worth it on a laptop.

Sourcing. Figures on this page are either derived from published engine or hardware constants, or computed from the formulas shown. Manufacturer-stated specifications are labelled as such. Found an error? Send a correction.