How Frame Generation Alters Traditional FPS Benchmarking

Frame generation technology has fundamentally disrupted the decades-old standard of using Frames Per Second (FPS) as the ultimate metric for computer hardware benchmarking. By using artificial intelligence and optical flow accelerators to insert synthetic frames between traditionally rendered ones, technologies like NVIDIA DLSS 3, AMD FSR 3, and Intel XeSS artificially inflate frame rates without proportionally increasing true rendering performance. This article explores how frame generation breaks the historical link between high FPS and low input latency, complicates direct GPU comparisons, introduces image quality variables into performance testing, and forces hardware reviewers to adopt entirely new benchmarking methodologies.

Synthetic Frames vs. Rendered Frames

Traditionally, an FPS counter measured how many frames the CPU and GPU could simulate and draw from scratch every second. A higher number meant the hardware was objectively faster at processing game logic and geometry.

Frame generation changes this by analyzing two completed frames and generating an intermediate "fake" frame without involving the CPU or the game engine. If a benchmark reports 120 FPS with frame generation enabled, the hardware is only truly rendering 60 frames per second. Consequently, comparing a graphics card running native rendering to one using frame generation is no longer an apples-to-apples comparison of raw compute power.

The Disconnect Between FPS and Input Latency

Before frame generation, higher FPS always resulted in lower input latency. A game running at 120 FPS felt twice as responsive to mouse and keyboard inputs as a game running at 60 FPS.

Frame generation severs this connection. To create an interpolated frame, the GPU must hold back a newly rendered frame to compare it with the previous one. This delay adds input lag. Furthermore, the generated frames do not process user input. As a result, a game running at 120 FPS with frame generation will look as smooth as a native 120 FPS experience, but it will feel like a 60 FPS (or worse) experience in terms of responsiveness. Because of this, hardware benchmarks can no longer rely on FPS alone to indicate how a game feels; they must now include end-to-end system latency measurements in milliseconds.

The Requirement for Baseline Performance

In traditional benchmarking, any boost to FPS was beneficial, especially at the low end. Pushing a struggling GPU from 25 FPS to 45 FPS via hardware overclocking or resolution scaling made an unplayable game playable.

Frame generation does not work well at low frame rates. Because the technology relies on existing frames to guess what happens next, a low base frame rate means there is too much time—and too much visual change—between frames. Using frame generation to boost 20 FPS to 40 FPS results in severe visual artifacts and sluggish, unplayable input lag. Frame generation requires a baseline of roughly 60 native FPS to function correctly. Therefore, benchmarks cannot use frame generation to crown a budget GPU as viable for heavy workloads, as the technology is primarily a "win-more" feature for hardware that is already performing well.

Image Quality as a Performance Metric

Historically, a rendered frame at a set resolution looked identical regardless of the GPU brand producing it, making FPS a purely quantitative metric. Frame generation introduces qualitative variables into benchmarking.

Because generated frames are only estimates, they often contain visual errors, such as ghosting behind fast-moving objects, shimmering, or garbled heads-up display (HUD) elements. Furthermore, one vendor's frame generation algorithm may produce fewer visual flaws than another's. Hardware testers can no longer just look at a bar chart of frame rates; they must visually inspect the output to determine if the artificially high FPS comes at the cost of a degraded image.

How Hardware Benchmarking Has Adapted

To maintain accuracy and consumer trust, modern hardware benchmarking has evolved to accommodate frame generation through the following changes:

  • Segregated Charts: Reviewers no longer lump all results into one graph. Benchmarks are now strictly separated into Native Rendering, Upscaling (DLSS/FSR Super Resolution), and Upscaling + Frame Generation categories.
  • Mandatory Latency Testing: Tools like NVIDIA FrameView and hardware latency analyzers are now standard in benchmarking suites to show the latency penalty incurred by turning on frame generation.
  • Focus on Frame Pacing: An average FPS counter can hide micro-stutters. Reviewers now look closely at frame time graphs to ensure the generated frames are spaced evenly between real frames, as poor frame pacing ruins the illusion of smoothness.
  • CPU Bottleneck Masking: Because frame generation bypasses the CPU, it can artificially mask a weak processor in GPU benchmarks. Testers must disable frame generation to accurately measure CPU gaming performance.