DLSS 4.5 Dynamic Frame Generation: Why Your FPS Counter Lies Harder Now
Generated frames do not advance the game state, so your input latency tracks the internal render rate, not the number on the FPS counter. DLSS 4.5 added Dynamic Multi Frame Generation, which varies the multiplier at runtime — so the counter is now not just inflated but inconsistently inflated.
What the counter is actually counting
Frame generation inserts AI-generated frames between rendered ones. Those inserted frames are shown to you, so they count toward presented frames per second, but they contain no new game state. Your input was not sampled for them and the simulation did not advance.
The practical consequence is that a game reporting 238fps with multi-frame generation is not doing what a game rendering 238 discrete frames does. Render internally at 40fps with a 4x multiplier and you get a smooth-looking 160fps display and the responsiveness of 40fps.
The latency numbers
NVIDIA Reflex is mandatory when frame generation is active, and it claws back a meaningful amount of CPU-side latency. Published measurements put the remaining penalty at roughly 8 to 12 milliseconds in most games.
One genuinely useful detail: that penalty does not grow as you raise the multiplier from 2x to 4x. The cost is largely in enabling generation at all, not in how many frames you generate. If you have decided to use it, using more of it is close to free in latency terms.
Put 8 to 12ms in context. It is comparable to the difference between a 144Hz and a 240Hz display, and larger than the gap between 1000Hz and 8000Hz mouse polling. If you are spending money to reduce input lag, turning frame generation off in competitive titles is free and larger than most hardware upgrades.
Where the threshold sits
Below roughly 50 to 60fps before generation, the added latency becomes noticeable. That is the number worth remembering, because it inverts the intuition: frame generation helps least exactly where the frame rate is worst and you most want help.
The useful split is by genre. Single-player games on a high-refresh display are the ideal case — smoothness matters, a 10ms latency cost does not. Competitive shooters are the wrong case, because the internal render rate is what decides whether you win a duel.
Measure it yourself
You can separate the two numbers without any special tooling. The FPS test measures presented frames independently of what any game reports, and the reaction time test gives you a whole-chain figure that includes your display and input device.
For the underlying relationship between the two numbers, refresh rate vs frame rate explains why they are separate measurements, and what counts as a good reaction time gives you benchmarks to compare your own result against.
Sources: NVIDIA — Reflex · MDN — requestAnimationFrame()
Frequently asked questions
Why does my FPS counter read high but the game feel sluggish?
Generated frames are shown to you but contain no new game state — your input was not sampled for them and the simulation did not advance. Render internally at 40fps with a 4x multiplier and you get a smooth-looking 160fps display with the responsiveness of 40fps.
How much latency does frame generation actually add?
Published measurements put the remaining penalty at roughly 8 to 12 milliseconds in most games, with NVIDIA Reflex mandatory and clawing back a meaningful amount of CPU-side latency.
Does a higher multiplier add more latency?
No. The penalty does not grow as you raise the multiplier from 2x to 4x — the cost is largely in enabling generation at all, not in how many frames you generate.
When should I turn frame generation off?
Below roughly 50 to 60fps before generation, where the added latency becomes noticeable, and in competitive shooters, where the internal render rate is what decides whether you win a duel.