The Witcher 3 Remastered Averages 161 FPS. The Same Run Reports 66 ms of Latency Nobody Mentioned

By Indie Kings | September 28, 2026

Updated September 28, 2026: Exactly one hard performance measurement for The Witcher 3 Remastered exists outside the developer, and it has been reported as a single number. It is not a single number. The full dataset from a pre-release build, on an RTX 5090 with a Ryzen 7 9800X3D at 4K with everything maximum and frame generation enabled, is 161.7 FPS average. The same run reports 66.326 ms average PC latency. That second figure is roughly four frames at 60 Hz, it is the cost of the frame generation that produced the first, and for a game built on parries and dodge windows it matters more than the frame rate does. Nobody has reported it.

Image: The Remastered on PC. Credit: Wccftech.

The full dataset, all ten metrics

Every figure below comes from one custom in-game test in one location. None of them is an average across a play session, and none of them is from a machine resembling a purchasable configuration.

One of those rows deserves an immediate note so it is not read as an error. The 1% low reads 131 while the minimum frame rate reads 135, which looks backwards. It is not a mistake: the minimum is a single frame, while the 1% low is a percentile average over the slowest 1 percent of the run, so on a short custom test one unusually slow frame can sit below the percentile average without contradicting it. Anyone comparing this table against a benchmark that reports a minimum should expect the same relationship and should not treat it as an error in either.

What was actually tested, and where

The configuration matters as much as the numbers, and it is a specification nobody would buy.

• Graphics card: GeForce RTX 5090. The requirements table's top tier asks for an RTX 5080, so this is one tier above the official ceiling.
• Processor: Ryzen 7 9800X3D. Again the requirements table's top choice, for path tracing.
• Every graphics setting set to the maximum available option.
• DLSS Super Resolution set to Quality, DLSS Multi Frame Generation set to Dynamic.
• 4K resolution.
• Build version 5.00, tested on a pre-release build ahead of the embargo with, in the reviewer's words, code that came in very hot over the weekend.

The location is specific and the reason is sound. The test was a custom route in Novigrad, from the docks to the furthest point in the city, chosen because that is the largest available area in the game and also the busiest in terms of NPC density. That is the correct choice for a path traced test, because dense crowds are what stress a renderer and what expose stutter, and a test run in open countryside would flatter the numbers.

So the methodology is better than most published benchmarks get credit for. A named build version, a named route with a stated reason for choosing it, a full settings list, and ten metrics including stuttering and latency is a genuinely rigorous capture. The problem is not the quality of the measurement. The problem is that one route on one machine is one route on one machine.

66 milliseconds, and why this particular game punishes it

This is the finding, and it is arithmetic before it is opinion.

At 60 Hz, a frame is about 16.7 milliseconds. An average PC latency of 66.326 milliseconds is roughly four frames of input delay at that refresh rate, and it is on top of whatever the display and the game already contribute. At 120 Hz, where a frame is 8.3 milliseconds, it is closer to eight.

Where the latency comes from is the frame generation, and the configuration tells the story plainly. DLSS Multi Frame Generation was set to Dynamic, which synthesises additional frames from the frames the GPU actually renders. That is what turns a plausible base rate into 161.7. Synthesised frames are generated rather than rendered, and generating them sits between an input and the frame that displays it, which is where the delay accumulates.

Now put that against what The Witcher 3 asks a player to do. Combat here is about reading an enemy's wind-up, dodging at the right moment, parrying, and casting signs in sequences. A reviewer who tested the build concluded that hitting enemies still does not feel punchy and that the lack of proper hit feedback is one of the reasons, which is a separate criticism but the same category of observation. This is a game where the gap between an input landing and a response arriving is the difference between a successful dodge and a corpse. Adding four frames to that is not a graph shape, it is a different game.

And here is the part that makes the configuration arguably wrong for the hardware it ran on. An RTX 5090 averaging 161.7 FPS with frame generation enabled has roughly two and a half times the frames it needs at 60 Hz before synthesis is involved. The 66 milliseconds is not the price of needing frame generation. It is the price of using it on a machine that was already going to be far above target without it. Run the same settings with Multi Frame Generation off, on that card, at 4K path traced, and the honest expectation is a frame rate still comfortably above 60 with a fraction of the latency.

So the reported headline is a number produced by a feature that a machine of that class does not need, delivering a latency penalty that a game with this combat model cannot afford. That is not a criticism of the reviewer's capture, which is careful and well documented. It is a criticism of reading 161.7 FPS as the answer to "how fast is this game."

The 1% low and the 0.1% low, which the site has never once reported

A vocabulary note, because it is the actual gap here rather than a technicality.

Across this site's entire archive, of more than eighteen thousand posts, the terms 1% low and 0.1% low do not appear. Neither does any equivalent percentile metric. Average frame rate appears constantly, and it is the weakest number in the table above.

The reason to prefer the percentiles is that a frame rate average is a statement about a machine, while a 1% low is a statement about an experience. A game can average 160 and feel like 100 in the moments that matter, because the moments that matter are the worst moments and the average does not describe them. On this run the gap between the average and the 0.1% low is 44.7 FPS, which on a 60 Hz display is unnoticeable and on a 144 Hz display is the difference between comfortably inside the refresh window and dipping below it.

The 6.185 millisecond average frame time is the number to hold onto if only one is remembered, because at 60 Hz the budget is 16.7 milliseconds and this run used about 37 percent of it before frame generation. The 0.573 millisecond variability is the wobble around that, and the 0.110 milliseconds of stuttering is the hitching that frame rate averages are designed to hide.

What this says about the official hardware tiers

Cross-referenced against the requirements table published the same day, this run is informative in a specific way.

The official Path Tracing tier asks for an RTX 5080 with a Core i9-13900K or a Ryzen 7 9800X3D, 16 GB of VRAM and 24 GB of RAM, targeting 2160p at 60 FPS with frame generation. This test used a card one tier above that and a CPU matching the top AMD option, at the same 4K class of resolution the official tier also targets, and averaged 161.7.

So the official top tier is conservative by a wide margin on a card one step up. That is not unusual or damning. Published targets are chosen with margin for the worst case and for the wide range of systems at that tier, and a single Novigrad route is not a worst case. What it does establish is that the target is not aspirational in the sense of being unreachable, which is the useful thing to know before buying a 5080 on the strength of the table.

The more interesting consequence is about the frame generation target itself. The official tier asks for 60 FPS with frame generation, which means the developer's expectation is that the base frame rate on a 5080 sits meaningfully below 60 and is being topped up. A player buying to that spec is buying a card that needs frame generation to hit its target, and on this game's combat that is the trade the official table is implicitly making. Whether it is a good trade depends entirely on the reader noticing that the tier is defined with frame generation on, which is stated in the table and is easy to skim past.

How it compares to the other two measurements that exist

Three measurements exist for this game. Two of them came from this site and one of those is the relevant comparison.

The Steam Deck pre-release measurement put the stock Deck preset at roughly 42 to 44 FPS after the remaster, down from 50 to 60 before, averaging about 44 with one dip to 39. That is the opposite end of the hardware spectrum from an RTX 5090, and between the two of them there is a usable shape: an eight gigabyte handheld losing about a quarter of its frame rate to the upgrade, and a top-tier desktop clearing 60 with room to spare.

What the Deck number lacks is the percentile and latency vocabulary, which is precisely why the 1% low and 0.1% low absence matters. A single observed dip to 39 FPS out of a 44 FPS average is exactly the kind of fact that percentiles would formalise, and it is the number a reader on a Deck actually needs. The three console targets, 30 fps quality and 60 fps performance on PS5 and Series X, and 30 to 40 fps on Switch 2, are developer targets with no measurement attached, so the gap between the two measured points is a wide one filled mostly with targets.

Set out plainly, the state of performance knowledge for The Witcher 3 Remastered as of September 28 is: one rigorous pre-release capture on hardware above the official ceiling, one pre-release handheld capture without percentile data, six developer target tiers for PC, two for PS5 and Series X, one for Series S, and two for Switch 2. Everything in between is interpolation.

What one pre-release run on one machine cannot tell you

The honest limits, listed rather than buried, because a performance article that omits them is the problem this site has been criticising elsewhere.

• One route, not a session. Novigrad's docks to its furthest point is dense and demanding. It is not Novigrad at night, not the open countryside, not a boss fight, and not a hundred hours of play.
• One machine, well above the official ceiling. An RTX 5090 is a tier above what the requirements table asks for at the top. This says nothing about a 5080, a 4070, a 1660, an Arc B580 or an RX 6600 XT, and nothing about the FSR 2 fallback that an RX 7000 or newer restriction imposes on the recommended AMD card.
• Pre-release build 5.00, and day-one patch 1.01 exists. The numbers may not be the numbers that ship.
• Frame generation on, latency unaddressed by the headline. The 66 milliseconds is reported, and the average is reported, and almost nobody connects them.
• No upscaler comparison. The run used DLSS Super Resolution on Quality. Nobody has published the same route with FSR 4, XeSS 2, or with the upscaler off, so the DLSS-specific margin is unknown.
• No path-tracing-off comparison. The run had path tracing on. The requirements table's RT Recommended tier is a different configuration and has no measurement at all.

FAQ

What frame rate does The Witcher 3 Remastered run at on PC?

The only rigorous measurement published is from a pre-release build on an RTX 5090 with a Ryzen 7 9800X3D, at 4K with every graphics setting maximum, DLSS Super Resolution on Quality and DLSS Multi Frame Generation on Dynamic, tested on a custom route across Novigrad. It averaged 161.7 FPS, with a 1% low of 131 and a 0.1% low of 117, an average frame time of 6.185 ms and 0.110 ms of stuttering. No measurement exists for any other GPU, so there is no published frame rate for a 5080, 4070, 1660, Arc card or Radeon card. CD Projekt Red's official targets are 30 FPS at Minimum and 60 FPS at Recommended, Ultra, RT Recommended and Path Tracing with frame generation.

What are the 1% low and 0.1% low, and why do they matter more than the average?

The 1% low is the average frame rate of the slowest 1 percent of frames in a run, and the 0.1% low is the same for the slowest one frame in a thousand. They matter more because a frame rate average describes a machine while a percentile describes an experience. A game can average 160 FPS and feel like 100 during the moments that matter, because the moments that matter are the worst moments and the average does not describe them. On the measured run the gap between the 161.7 average and the 117 0.1% low is 44.7 FPS, which is unnoticeable at 60 Hz and the difference between staying inside and dipping below the budget at 144 Hz.

What is the 66 millisecond PC latency and should I care?

It is the input-to-display delay contributed by the pipeline, and on that measured run it is 66.326 ms average, roughly four frames at 60 Hz. It comes from frame generation, which synthesises frames rather than rendering them and therefore sits between your input and the frame you see. For a game built on reading enemy wind-ups, dodging, parrying and sign timing, four extra frames changes the game. The important part is that it is not a necessary cost on that hardware: an RTX 5090 averaging 161.7 FPS has roughly two and a half times the frames it needs at 60 before synthesis, so the latency is the price of using a feature that machine did not need rather than the price of needing it.

Does the official 5080 path tracing target match this measurement?

The test sits above the target rather than confirming it. The official Path Tracing tier asks for an RTX 5080, a Core i9-13900K or Ryzen 7 9800X3D, 16 GB of VRAM and 24 GB of RAM for 2160p at 60 FPS with frame generation, and this run used a card one tier above that at the same resolution class and averaged 161.7. A conservative target is normal, since published tiers are chosen with margin for worst cases rather than from one good route, so the useful conclusion is that the target is reachable rather than that the hardware is overkill. Note also that the official tier is defined with frame generation, which implies a base frame rate meaningfully below 60 being topped up.

What is the most useful number in the dataset?

The average frame time of 6.185 ms, because it is the one figure that can be compared against a display budget directly. At 60 Hz the budget is 16.7 ms, so this run used roughly 37 percent of it. The 0.573 ms of FPS variability is the wobble around that figure, and the 0.110 ms of stuttering is the hitching that a frame rate average is specifically designed to conceal. If you only read one row of that table, read the frame time and the stuttering rather than the average.

How does that compare to the Steam Deck and the consoles?

It is the opposite end of the hardware spectrum from the only other measurement that exists. The Steam Deck pre-release test found the stock preset falling from 50 to 60 FPS before the remaster to roughly 42 to 44 after, averaging near 44 with one dip to 39, and it reports no percentiles at all. The console figures are developer targets with no measurement behind them: 30 FPS quality and 60 FPS performance on PS5 and Xbox Series X, 30 to 40 on Switch 2, and 1440p at 30 or 1080p at 60 on Series S. So the honest state of knowledge is one rigorous capture on hardware above the official ceiling, one handheld capture without percentile data, and a set of targets filling the space between. Everything in that middle is interpolation.

Bottom Line

The reason this matters more than a fourth preview is that almost every number a reader will find about this game's performance today is a target or a single average, and the one real measurement that exists has had its interesting half discarded. An RTX 5090 with a Ryzen 7 9800X3D at 4K, everything maximum, averaging 161.7 FPS, is a genuinely impressive result and it is also the least informative figure in the dataset. The run reports ten metrics. Five of them describe how the machine performed and one describes how the game will feel, and that one is 66.326 milliseconds of average PC latency.

The latency is the cost of the frame generation that produced the headline. Multi Frame Generation synthesises frames rather than rendering them, which places a delay between an input and the frame displaying it, and at 60 Hz that 66 milliseconds is about four frames. The Witcher 3 asks players to read enemy wind-ups, dodge on the timing, parry, and chain signs, so the distance between an input and a response is the difference between a successful dodge and a corpse. A reviewer who tested the same build separately concluded that hits still lack proper feedback, which is a different criticism in the same family. None of that is visible in 161.7.

The sharper version of the point is that the configuration is arguably wrong for the hardware. A card averaging 161.7 FPS with frame generation on has about two and a half times the frames it needs at 60 before synthesis is involved, so the 66 milliseconds is not the price of needing frame generation. It is the price of switching on a feature that machine was already far past needing. The same card, same settings, same route with Multi Frame Generation off would be expected to hold comfortably above 60 with a fraction of the delay. The capture is careful and well documented, with a named build version, a named route chosen for the right reason, and a full settings list, so this is not a criticism of the review. It is a criticism of reading an average as the answer to how fast a game is.

There is also a genuine vocabulary gap on this site, and it is the concrete reason the interesting half was discarded. Across more than eighteen thousand posts, the terms 1% low and 0.1% low do not appear once, and neither does any equivalent percentile. Average frame rate appears constantly and it is the weakest number available, because an average describes a machine while a percentile describes an experience. This run's gap between a 161.7 average and a 117 0.1% low is 44.7 FPS, which nobody would feel at 60 Hz and which is the difference between staying inside and dipping below the frame budget at 144 Hz. The single most useful number in the table is the 6.185 millisecond average frame time, because at a 16.7 millisecond budget for 60 Hz it can be compared directly, and the 0.110 milliseconds of stuttering is the thing players actually notice and averages are built to hide.

Read against the official tiers, the measurement says the top target is reachable rather than aspirational. The requirements table asks for an RTX 5080 with a 9800X3D or 13900K for 2160p at 60 with frame generation, and this run is a card one tier up at the same resolution class averaging 161.7. That gap is normal, because published tiers are chosen with margin rather than from one good route in one city. The consequence worth noticing is that the official top tier is defined with frame generation on, which implies the base frame rate on a 5080 sits meaningfully below 60 and is being topped up. A buyer reading that table is buying a card that needs frame generation to reach its target, on a game whose combat model punishes the latency that feature adds. It is stated in the table and easy to skim past.

Set against the other two measurements that exist, the state of knowledge is thinner than the volume of coverage suggests. The Steam Deck pre-release test found the stock preset dropping from 50 to 60 FPS to roughly 42 to 44 after the upgrade, averaging near 44 with one dip to 39, and it reports no percentiles at all, which is exactly where the 1% low gap bites because a single dip to 39 out of a 44 average is the fact a Deck owner needs and the fact averages conceal. The console figures, 30 and 60 on PS5 and Series X, 30 and 40 on Switch 2, are targets with no measurement behind them. So: one rigorous capture on hardware above the official ceiling, one handheld capture without percentile data, and a set of targets filling the space between. Everything in that middle is interpolation, and this article does not pretend otherwise.

Six limits are worth stating plainly, because a performance article that omits them is the failure this site criticises elsewhere. One route in one city is not a session. One machine well above the official ceiling says nothing about a 5080, a 4070, a 1660, an Arc card or a Radeon, and nothing about the FSR 2 fallback that the RX 7000-or-newer restriction imposes on the recommended AMD card. The build is pre-release version 5.00 and day-one patch 1.01 exists, so the numbers may not be the numbers that ship. There is no comparison with the upscaler off, or with FSR 4 or XeSS 2, so the DLSS-specific margin is unknown. And there is no path-tracing-off comparison, so the RT Recommended tier has no measurement at all. What the dataset does establish is a high ceiling, a clean stutter profile at 0.110 milliseconds, and a latency figure that the game's combat specifically cannot absorb. All three are worth more than the average that was reported instead of them.

Source: Wccftech, The Witcher 3: Wild Hunt Remastered Looks Incredible With Path Tracing, but the Combat and Characters Still Show Its Age by Alessio Palumbo, testing pre-release build version 5.00, is the primary and sole source for every measured figure in this article. That includes all ten metrics of the custom test, the average of 161.7 FPS, the 1% low of 131, the 0.1% low of 117, the minimum of 135, the maximum of 199, the average frame time of 6.185 ms, the FPS variability of 0.573 ms, the 0.110 ms of stuttering, the comparable frame rate of 145.6 FPS and the average PC latency of 66.326 ms, together with the full test configuration of an RTX 5090, a Ryzen 7 9800X3D, every graphics setting at maximum, DLSS Super Resolution on Quality, DLSS Multi Frame Generation on Dynamic, and 4K resolution, and the location and rationale of the custom Novigrad route from the docks to the furthest point of the city as the largest and most NPC-dense area available. That article is also the source for the observations that hits still lack proper hit feedback, that character models improved only slightly, that the water looks worse than everything else and was not improved, and that the path traced hair option did little. CD Projekt Red's updated PC requirements support article of September 28, 2026 is the source for the official tiers, including the Path Tracing tier of an RTX 5080 with a Core i9-13900K or Ryzen 7 9800X3D at 2160p and 60 FPS with frame generation, and for the 30 FPS Minimum and 60 FPS Recommended, Ultra, RT Recommended targets. This site's Steam Deck pre-release measurement is the source for the Deck figures of roughly 42 to 44 FPS after the remaster against 50 to 60 before, averaging near 44 with one dip to 39, and this site's earlier coverage of the Switch 2 and console targets is the source for those figures. The conversion of 66.326 ms into approximately four frames at 60 Hz, the reading that the configuration's latency is the price of an unnecessary feature rather than of a needed one, the observation that the 1% low and 0.1% low terms appear nowhere in this site's archive, the framing of the 6.185 ms frame time as the most directly comparable number in the set, and the observation that the official top tier is defined with frame generation enabled are this article's. No performance figure is presented for any GPU other than the one measured, and no number is estimated or extrapolated

Related: Witcher 3 Remastered Sept 29: Free Upgrade, 30 Changes [2026] | Switch 2 File Sizes Led by Witcher 3 Remastered at 43.5GB [2026] | Witcher 3 Remastered: Free Upgrade, Sept 29 Launch [2026] | Intel Driver 101.8531: 35% Witcher 3 FPS Boost on Core Ultra 3 | Witcher 3 Remastered Switch 2 Locks Sept 29: 42GB, 40FPS, Free Upgrade [2026]