Enable in-game NVIDIA Reflex first when it is available. If a game has no Reflex option, try Low Latency Mode on Ultra in that game’s NVIDIA Control Panel profile, then keep it only if responsiveness improves without worse frame pacing.
I would leave the global setting at Off and make exceptions per game. That gives you a clear starting point and makes an unhelpful change easy to undo. On is also worth comparing if Ultra feels uneven; the most aggressive setting does not automatically produce the best experience.
01What Off, On, and Ultra actually change
Low Latency Mode controls how much rendering work the CPU can prepare ahead of the GPU. NVIDIA describes Off as prioritizing throughput, On as limiting queued frames to one, and Ultra as fully minimizing the queue. Those are scheduling choices, not graphics-quality presets.
| Setting | Queue behavior | My use |
|---|---|---|
| Off | Allows game frame queuing. | Baseline and Reflex games. |
| On | Limits queued frames to one. | Alternative to compare if Ultra feels worse. |
| Ultra | Minimizes queued frames. | First latency-focused trial without Reflex. |
Do not read Ultra as “zero total input lag.” Even with little waiting in the queue, the game still has work to perform. A useful decision asks whether that specific waiting stage is hurting your experience, rather than whether every available performance switch is enabled.
02Use Reflex before the driver setting
Reflex is integrated into supported games and coordinates work inside the engine. The driver setting operates from outside it. NVIDIA recommends Reflex over the driver’s Ultra mode, and Reflex takes priority if both are enabled.
My recommendation is to enable Reflex in the game, leave the driver setting Off for that profile, and evaluate that configuration first. This keeps the setup understandable. Switching the driver option on as well does not establish an additional, stackable latency reduction.
Start with ordinary Reflex On. On + Boost is a separate decision: it can keep GPU clocks elevated when the GPU is underused, with a power trade-off. I would compare Boost only after the basic setup is satisfactory, especially on a laptop where fan noise matters.
03DirectX 12 advice needs the driver date
The claim that NVIDIA’s driver low-latency option cannot work in DirectX 12 is outdated. NVIDIA’s 551.23 release, dated January 24, 2024, explicitly added Ultra Low Latency Mode support for DirectX 12 titles.
That is a support milestone, not a recommendation to install an old driver. Use a suitable current driver for your GPU and operating system. The practical result still needs checking in your actual game; API support alone does not predict a measurable improvement.
Check Reflex first. Without it, compare the driver option using the same renderer, scene, and graphics settings.
Do not extend the DX12 support announcement to Vulkan. Prefer a documented in-game latency feature; do not assume this driver switch controls its queue.
I would not switch a working game between DirectX and Vulkan just to make this option relevant. Changing the renderer introduces another variable and may alter the entire comparison. First establish what the game already supports.
04When I would try it—and when I would stop
The strongest reason to experiment is a game that feels delayed despite otherwise consistent motion. A workload where the GPU cannot keep up with prepared frames is a plausible place for queue management to matter. A CPU-limited game can respond differently.
Treat that as a reason to test, not a diagnosis from one utilization reading. I would choose a representative gameplay section rather than a loading screen, menu, or unusually empty room. Keep the same frame cap while comparing modes.
You care about aiming or steering response, the game lacks Reflex, and you can reproduce the same scene.
Ultra introduces a repeatable uneven feeling, lowers performance unacceptably, or produces no useful difference.
For a relaxed single-player game already meeting your needs, leaving the setting Off is reasonable. My priority would be consistent control and motion during ordinary play. A mode name is not evidence that your configuration has improved.
05Change only the game you want to improve
- Record the starting point. Note the game’s Reflex, V-Sync, frame-cap, and rendering settings before changing anything.
- Open NVIDIA Control Panel. Go to Manage 3D Settings, then Program Settings.
- Select the actual game executable. Use Add or Browse if needed. Avoid selecting only a launcher when the game runs through another executable.
- Find Low Latency Mode. Without in-game Reflex, select Ultra for the initial comparison and click Apply.
- Relaunch the game. Revisit the same scene and confirm the intended profile still shows the selected value.
For a precise reference to the options, NVIDIA’s driver settings reference distinguishes global defaults from program-specific profiles. I prefer the latter because it makes each game’s configuration easier to explain and reverse.
If the expected control is missing, confirm which GPU renders the game and whether the installed NVIDIA driver recognizes it correctly. I would investigate that mismatch before using third-party profile tools to force hidden values.
06Keep G-SYNC, V-Sync, and caps in the comparison
Decide whether your priority is the lowest possible delay or a tear-free picture. NVIDIA’s latency guidance favors V-Sync Off when tearing is acceptable. It also describes G-SYNC with V-Sync and Reflex or Ultra as a route to keeping FPS below the refresh ceiling in supported configurations.
I would verify the actual frame rate instead of assuming an automatic cap is active everywhere. Display mode, the game, and laptop display routing can affect which controls apply. If your existing cap already gives the result you want, keep it unchanged for the first comparison.
A fixed-refresh 60 Hz monitor is not a G-SYNC monitor simply because it is connected to an NVIDIA GPU. Choose settings for the display you actually use. Do not copy a high-refresh competitive configuration without checking what problem its synchronization choices solve.
Change synchronization separately from Low Latency Mode. If you disable V-Sync, remove a cap, and enable Ultra together, any improvement will be difficult to attribute. Once you choose a driver mode, you can run a second comparison focused on synchronization.
07Compare response and frame delivery separately
Use the same saved-game location, training sequence, or repeatable camera movement. My suggested routine is several comparable passes with Off, then Ultra, then Off again. Keep resolution, upscaling, recording software, frame generation, and the frame cap unchanged.
Watch both the response to an action and the consistency of motion. If your game exposes a latency measurement, record its exact label. Render latency and click-to-display latency describe different portions of the system; do not compare them as though they were interchangeable.
Without a suitable measurement, describe a subjective improvement as a preference. Do not turn “aiming feels better” into a claimed millisecond reduction. Likewise, a higher FPS counter alone does not establish that input-to-display delay fell.
If Ultra looks worse, compare On against the original Off configuration. If the result is inconsistent across runs, keep the simpler baseline and revisit only when you have a repeatable symptom. There is no obligation to find a benefit.
08Separate queue tuning from other performance problems
I would investigate a pause triggered by entering the same area separately from a generally delayed control response. Keep notes about where it happens, whether it repeats, and whether the symptom changes after the driver setting is restored.
The distinction between shader and traversal stutter provides a useful supporting workflow when an isolated hitch is the real complaint. Avoid changing caches and latency controls in the same experiment.
Similarly, leave GPU clock tuning alone while evaluating this setting. Cubetoons’ NVIDIA Automatic Tuning guide covers that separate decision. Keeping those experiments apart lets you identify which change helped or caused a regression.
If the game is satisfactory after returning to Off, keep that working configuration. If the problem remains, follow the symptom rather than repeatedly cycling the same three modes.
09Questions before you leave it enabled
Should I enable it globally?
I would keep the global default at Off and use individual profiles. You can then see exactly which games have an exception, and a disappointing experiment does not alter unrelated applications.
Is On always safer than Ultra?
Both are ordinary driver options. The useful question is which produces the better result in your game. Compare On if Ultra is unsatisfactory; do not infer a universal ranking from the labels.
Can I leave both Reflex and Ultra enabled?
Reflex takes priority. I would leave the driver option Off in a Reflex profile so that the setting responsible for latency control is clear.
How do I undo the change?
Return to the same Program Settings profile, restore the recorded Low Latency Mode value, and apply it. Avoid resetting the entire profile if you want to preserve other deliberate settings.
10My recommendation: decide one game at a time
Start by checking the in-game Reflex option. Without Reflex, give Ultra one controlled per-game trial and compare On only if needed. Keep whichever configuration produces repeatable, useful responsiveness with acceptable frame delivery.
The next action is small: choose one game, record its baseline, and compare the same scene. Leave the rest of your library alone until you have a reason to change it.