On October 13, 2026, NVIDIA announced an update that makes the DLSS 5 AI model twice as fast as the September version. The VRAM used by the DLSS 5 model also drops by 20% at 4K, and the update arrives in NBA 2K27 on October 16. The feature that began in that game in September, which uses AI to enhance lighting and material appearance, will now run with less overhead.
However, the 2x figure refers to the model's processing speed, which is a different measure from the game's overall frame rate. Hogwarts Legacy and Where Winds Meet are also scheduled to receive official support in November. To judge this optimization, we need to look at how the GPU time and memory spent on the added image quality have changed.
What the "2x faster" covers, and what the 20% VRAM cut applies to
According to NVIDIA, the October model has 24% fewer parameters, down to roughly 118 million. Combining overall pipeline optimization with model improvements, the company says the neural processing load on the GPU is halved while image quality stays comparable to the September version. Compared with the March debut, NVIDIA says the model's processing speed has reached 10x.
Meanwhile, the in-game frame rate improvement NVIDIA cites for NBA 2K27 is up to 1.4x. That figure compares the new model with the old one. It does not mean the game runs faster than it does with DLSS 5 turned off.
The "2x," "20% reduction," and "up to 35%" figures each measure a different thing: neural processing speed, VRAM at 4K, and frame generation, respectively.
| Improvement announced by NVIDIA | What the figure refers to | Caveats on comparison and interpretation |
|---|---|---|
| Model processing speed 2x | DLSS 5 neural processing, compared with the September version | Does not mean overall game fps doubles |
| In-game fps up to 1.4x | NBA 2K27 using the new model | Not a comparison against DLSS 5 turned off |
| VRAM use down 20% at 4K | Memory used by the DLSS 5 model | NVIDIA gives no overall game VRAM reduction rate or absolute GB savings |
| Up to 35% faster | New frame generation mode on the RTX 5060 | An improvement to a separate feature, with a small image quality trade-off |
The source is NVIDIA's October 13 announcement. The table lists each feature's improvement by target; these are not independent benchmarks measured on the same GPU and settings. The numbers also cannot be added or multiplied to derive an overall game speedup.
The VRAM figure in particular needs careful reading. GPU memory also holds game textures and rendering data. Even if the memory tied to the model shrinks, those other items do not shrink by 20% with it. NVIDIA's update notes give no before-and-after absolute amounts, so this number alone cannot tell you whether your GPU now has enough headroom.
The image quality DLSS 5 adds has a computational cost
DLSS 5 neural rendering takes the image drawn by the game engine as a base and adds lighting and material detail. Its job differs from super resolution, which reconstructs a low-resolution image at a higher resolution, and from frame generation, which creates intermediate images between rendered ones. A step that alters image quality is added to the end of the conventional rendering pipeline.
NVIDIA's technical explanation describes a design that takes one frame as input and outputs one frame, using the image's color and motion vectors, which indicate the direction and amount of motion. It aims to add detail such as light scattering within skin, light passing through hair, and shadows where objects meet, while keeping results stable in moving scenes.
In other words, even if neural processing gets faster, the time the game spends drawing the original image remains. In a simple case where AI processing is added after rendering, total time is "original rendering time + AI processing time." Halving the second part does not halve the first. This difference in scope is why a 2x model speedup sits alongside an in-game fps gain of up to 1.4x. On real GPUs, how the workloads overlap also matters, so individual fps figures cannot be back-calculated from this explanation.
Similarly, the memory savings depend on how much of the game's total usage the neural processing accounts for. A smaller model is welcome, but parameter count, GPU processing time, and VRAM are different quantities. Nothing in this announcement justifies explaining the halved processing load by the 24% parameter reduction alone, or inferring that a particular compression technique was used.
DLSS 5 can be combined with other features such as super resolution and multi frame generation as an optional, independent feature. What players want to know is how much the cost of the image quality improvement has fallen compared with before the update. Isolating that cost requires comparisons with the other DLSS features held constant.
Hogwarts Legacy's official support will test control over the look
Hogwarts Legacy is scheduled to support DLSS 5 on November 5 for GeForce RTX 50 series, RTX Spark, and GeForce NOW. The same update switches the existing multi frame generation and ray reconstruction to their latest versions. Where Winds Meet follows in late November, adding neural rendering to the super resolution and multi frame generation it already supports.
NVIDIA explains that Avalanche Software used the DLSS 5 development kit to tune the effect and integrated it to preserve Hogwarts Legacy's intended art direction and style. Official support means more than the option appearing in the game's settings; it also means the developer chooses an application method that suits the title.
Developers can choose among multiple models and use them differently by scene. "Structure expression intensity" adjusts fine shading and reflections, while "tone expression intensity" adjusts broad lighting and color changes. Semantic-classification-based masks let developers apply the effect differently to characters and backgrounds, and engine-side masks can target specific objects. Use is not limited to applying one uniform strength across the whole screen.
Still, being able to adjust is not the same as looking as intended in every scene. Did improving the texture of faces and clothing in conversations also change the game's soft color tones? Do moving hair and outlines flicker? The November official launch will be a chance to check how well image quality and artistic style are balanced.
The quality of the source image also affects the result. NVIDIA explains that feeding richer lighting information through ray tracing or path tracing makes the neural processing more accurate. The design is not one where you can render coarsely and leave it to the AI to reach the same finish. Even after the model gets lighter, the balance between the compute spent on the original rendering and the expression added by AI will vary from title to title.
For RTX 40, what arrives first is a lighter frame generation
What RTX 40 series users can try this time is a fast frame generation mode in the beta version of the NVIDIA app. The setting is called "Preset C." It also applies to the RTX 50 series and RTX Spark, and NVIDIA touts up to a 35% speedup on the RTX 5060 while stating that there is a small image quality trade-off.
Conventional frame generation uses an AI model that understands motion from the game's depth and motion vectors, and another AI model that computes optical flow, the on-screen motion. The latter also handles elements such as UI and ray tracing effects. Preset C replaces this second model with a faster, lighter algorithm.
This is a different change from the lightening of DLSS 5, which generates lighting and materials. That this beta feature runs on RTX 40 does not mean DLSS 5 neural rendering or multi frame generation has received official support there. Also, the 35% figure is for the RTX 5060, and NVIDIA does not promise the same improvement on every RTX 40 product.
In its official FAQ, updated September 3, NVIDIA laid out a plan to tune RTX 50 performance sufficiently first, then extend official DLSS 5 support to RTX 40. The October update shows progress on optimization, but no release date for the RTX 40 version has been given this time either. When trying it on your own GPU, you need to check the frame generation setting and the neural rendering setting separately.
How far will the optimization spread through RTX Remix?
On October 23, the new DLSS 5 model and fast frame generation mode are also scheduled to arrive in RTX Remix, the tool for modding older PC games and overhauling their rendering. Fallout 4 RTX and Mirror's Edge RTX, which NVIDIA showed, are mod projects in development, not official updates to the original games.
Mod authors get the same controls as game developers. Using per-object settings and masks, they can choose which parts receive AI processing and which are excluded. By providing not only a faster model but also tools to tailor the effect to each work, NVIDIA widens the options for bringing it to existing games.
First, with the October 16 NBA 2K27 update, it will be worth comparing the September model, the new model, and neural processing turned off, on the same GPU and in the same scene. Resolution and quality settings should be matched, and super resolution and frame generation multipliers fixed. Beyond displayed fps including generated frames, look at how often the game actually renders and at moments when frame times temporarily lengthen. Adding input responsiveness and image quality in motion makes it possible to judge what kind of comfort the speedup brings.
Extending that comparison to November's officially supported titles will show how much of the time freed from AI processing can go toward rendering settings and preserving image quality. The condition for DLSS 5 to become an everyday game setting is gaining enough rendering headroom while preserving the expression the developer chose.