DLSS vs FSR 3 Latency Benchmark: Real FPS, Input & VRR Tested

A higher FPS counter can hide a slower game. That is the central finding of this DLSS vs FSR 3 latency benchmark: DLSS Frame Generation is usually the safer choice for latency and pacing on supported GeForce hardware, but FSR 3 can deliver excellent results—especially on Radeon—when the underlying frame rate is already healthy.

Key Takeaways

• Latency is not the same as displayed FPS: In ComputerBase’s Starfield test on an RTX 4070, DLSS Frame Generation measured 66.3 ms, while FSR 3 Frame Generation reached 83.3 ms despite producing more displayed frames.
• The architecture matters, but integration matters more: DLSS uses NVIDIA’s Optical Flow Accelerator, Tensor Cores and Reflex; FSR 3 uses a hardware-agnostic interpolation path with no Tensor Core requirement.
• Use frame generation above roughly 50–60 base FPS: Below 40 FPS, generated frames can make motion look smoother while controls still feel distinctly delayed.

What the technologies actually do

The naming causes unnecessary confusion. DLSS 3.5 is a broader NVIDIA feature generation that includes Super Resolution, Ray Reconstruction and Frame Generation. The direct comparison here is DLSS Frame Generation versus FSR 3 Frame Generation, both released into games during 2023.

Neither technology renders every displayed frame from scratch.

The game renders a sequence of real frames. The frame-generation system examines those frames, engine motion vectors, depth information and optical-flow data, then inserts an interpolated frame between them. That inserted frame can improve visual smoothness, but it doesn't contain a newly simulated mouse movement or game-state update.

That distinction explains why 120 displayed FPS doesn't feel identical to native 120 FPS. If the game is rendering only 60 real frames per second, the control response remains much closer to a 60-FPS pipeline than a native 120-FPS one.

DLSS has a more specialized hardware path. On supported RTX GPUs, optical flow is handled by NVIDIA's Optical Flow Accelerator, while Tensor Cores run the neural frame-generation model. NVIDIA also integrates Reflex to reduce queued frames and control the render pipeline.

FSR 3 doesn't require Tensor Cores or NVIDIA-specific hardware. AMD's implementation uses engine motion data and GPU-computed optical-flow analysis, which makes it available across a much wider range of hardware. That flexibility is useful, but it leaves more room for variation between games, drivers and GPU vendors.

The practical question isn't which optical-flow estimator sounds more advanced. It's this:

Does the complete pipeline deliver lower end-to-end latency, steadier frame times and fewer visible interpolation errors?

DLSS vs FSR 3 latency benchmark results

ComputerBase measured latency with LDAT in four games using an RTX 4070 and Radeon RX 7800 XT. The results are more revealing than a typical average-FPS chart because they show how differently frame generation can behave from one title to another.

Here are the most useful RTX 4070 results:

Those numbers don't produce a universal winner, but they do answer the question “does DLSS frame generation increase input lag?” Usually, yes, relative to the same upscaling mode without generation—but not always by the same amount.

In Like a Dragon, DLSS Frame Generation added approximately 4.9 ms over DLSS Performance. FSR 3 added about 6.5 ms over FSR Performance on the RTX 4070.

Starfield is the standout result. DLSS Frame Generation measured 66.3 ms, compared with 83.3 ms for FSR 3 on the same GPU. That's a 17 ms gap—large enough to feel during camera movement and aiming. FSR 3 may have reported the larger FPS gain in that scenario, but the extra displayed frames didn't translate into a faster control response.

Call of Duty produced a different result. Both frame-generation modes landed around 50 ms on the RTX 4070, far above the upscaling-only results. ComputerBase also found the FSR implementation felt poor in that particular title, despite apparently acceptable frame pacing and measured performance.

That last point matters. Measured latency, frame pacing and subjective control response are related, but they aren't interchangeable.

AMD's official FSR 3 figures show why the technology still has a strong use case:

In Forspoken, FSR 3 added 74 displayed FPS for roughly 7 ms of additional latency. That's a sensible trade in a demanding single-player game. AMD's numbers are vendor-supplied, so they shouldn't be treated as equivalent to an independent lab result, but they demonstrate the intended operating range.

The Radeon results also matter. In ComputerBase's Like a Dragon test, FSR Frame Generation on the RX 7800 XT measured 32.4 ms, slightly below the 33.5 ms result for FSR Performance without generation. That isn't a general law; it shows how strongly game integration and pipeline scheduling can affect the outcome.

Why FSR 3 can feel slower at the same FPS

The most common mistake is comparing displayed FPS without checking base FPS.

Suppose DLSS renders 60 real FPS and inserts enough frames to display 115 FPS. FSR 3 might render 52 real FPS but display 125 FPS. The FSR counter looks better, yet each input is still entering a slower underlying render loop.

Generated frames are visual predictions. They don't advance the simulation with a fresh input sample. If the base rate falls too low, the game can look fluid while the camera, crosshair and movement feel behind your hands.

That is why the best base FPS for frame generation is usually around 50–60 FPS, with 60 FPS or higher preferable for fast games. At 30–40 base FPS, frame generation can still help a slow-paced RPG or cinematic adventure, but it shouldn't be mistaken for native responsiveness.

There are other sources of delay:

• CPU render submission and game simulation time
• GPU queue depth
• Reflex, Anti-Lag or equivalent latency controls
• Swapchain and presentation behavior
• VRR state
• Monitor scanout time
• Frame-time variation between real and generated frames

A dedicated optical-flow accelerator can reduce shader workload and make the pipeline more predictable. It cannot erase those other delays. That is why “DLSS has hardware, FSR is software” is an incomplete explanation for benchmark results.

The technologies also differ in artifact behavior. HUD elements, thin foliage, particles, reflections and rapidly changing lighting are difficult because they may not follow the game's ordinary motion vectors. DLSS can use optical-flow information to account for motion in effects that conventional vectors miss, but it isn't artifact-proof. FSR 3 can look very clean in one game and produce distracting trails or UI errors in another.

DLSS and FSR 3 VRR frame pacing

Variable refresh rate complicates the FPS story further.

A 144 FPS result is of limited value on a 120 Hz monitor if generated frames repeatedly hit the refresh ceiling. Likewise, a game that falls below the display's VRR floor can stutter even while its average FPS looks impressive.

A useful test records more than an FPS counter:

For G-Sync or FreeSync, start with the display's correct VRR mode enabled and cap output slightly below the panel's maximum refresh rate. A three-FPS-under-refresh cap is a reasonable starting point, not a guaranteed answer. Some frame-generation implementations respond differently depending on the title and presentation path.

Test near three boundaries:

1. Just below the monitor's refresh ceiling.
2. Around the point where VRR becomes active.
3. At the lowest base FPS you consider playable.

If frame generation repeatedly pushes output above the VRR range, the display may fall back to fixed refresh behavior or introduce uneven presentation. If the base rate drops below the VRR floor, the result can feel rough even though the generated FPS remains high.

Public testing has not yet established a universal G-Sync-versus-FreeSync latency advantage for DLSS or FSR 3. Panel behavior, monitor firmware and game implementation make broad claims risky. The right approach is to measure refresh behavior on the exact display you use.

Which one should you use?

Choose DLSS Frame Generation when you have a supported RTX card, the game has a solid Reflex implementation, and you care about aiming, camera control or image stability. It is generally the safer option on GeForce hardware, particularly when the base frame rate sits above 50–60 FPS.

Choose FSR 3 Frame Generation when you use a Radeon GPU, need broader compatibility or the game produces a substantially higher base rate with FSR. On an RX 7800 XT, FSR 3 can be an effective way to make demanding single-player titles feel smoother. On an RTX 4070, don't assume it will match DLSS latency simply because it displays more frames.

Disable either mode for competitive shooters when the base rate is low, the frame pacing is uneven or the generated image shows obvious artifacts. A clean 75 real FPS is often preferable to a delayed 130 displayed FPS.

The useful rule is simple: judge frame generation by base FPS, latency, frame-time consistency and VRR behavior together. The largest number in the corner is only one part of the result.

Frequently Asked Questions

Q: Does DLSS Frame Generation increase input lag?

Yes, usually compared with the same DLSS upscaling mode without Frame Generation. The increase varies by game; ComputerBase measured roughly 4.9 ms in Like a Dragon on an RTX 4070, while an earlier Cyberpunk 2077 test on an RTX 4090 measured a larger increase from 71.2 to 86.3 ms. Reflex helps manage queueing but cannot make generated frames respond like independently rendered frames.

Q: Why does FSR 3 feel slower at the same FPS?

Displayed FPS includes interpolated frames that don't contain a new game simulation or input sample. If FSR 3 starts from a lower base FPS, controls can feel slower even when the output counter matches or exceeds DLSS. Game integration, queue depth, VRR behavior and artifacts can widen that subjective gap.

Q: What is the best base FPS for frame generation?

Aim for at least 50–60 real FPS before enabling it. Around 30–40 base FPS, frame generation can improve visual smoothness in slower games, but latency and uneven frame pacing are more likely to remain noticeable.

Q: Is FSR 3 Frame Generation worth using on an RTX 4070?

It can be, but test it per game. In ComputerBase's Starfield result, FSR 3 Frame Generation reached 83.3 ms versus 66.3 ms for DLSS Frame Generation on the same RTX 4070. If DLSS is available and delivers a similar base rate, it is usually the better latency choice.