Nvidia's DLSS 5 Gets a 2x Faster Model, and It Is Really About VRAM

Nvidia quietly swapped the brain inside DLSS 5 this week, and the spec sheet for the new model is the most aggressive efficiency play the company has made in AI upscaling yet. The 3D-Guided Neural Rendering model, which shipped with DLSS 5 in September 2026, is being replaced by a version that Nvidia says runs twice as fast, carries 24 percent fewer parameters, halves GPU overhead, and uses 20 percent less VRAM at 4K, all while holding visual fidelity roughly constant. It debuts in NBA 2K27 on October 16 and then rolls into RTX Remix, Hogwarts Legacy, and the new RTX Spark laptop line.

The announcement, which Nvidia posted October 13 under the title "DLSS 5 Comes To Hogwarts Legacy & RTX Spark, Powered By 2X Faster Model," is careful about one thing: this is not DLSS 5.1. There is no renumbered product. Nvidia is simply hot-swapping a more efficient neural network behind the existing DLSS 5 name, and future games that ship with DLSS 5 support will default to the new model going forward. That makes the October update less a one-time patch and more the new baseline for every DLSS 5 title released from here on.

What the numbers actually say

Strip the marketing language down and Nvidia is making four checkable claims, all attributed to its own testing. The new model runs twice as fast as the September version, meaning inference time is cut roughly in half. Its parameter count dropped by 24 percent to approximately 118 million, which is the technical reason it can run faster while demanding less video memory. GPU overhead from running the model is halved. And 4K VRAM usage falls by 20 percent compared to the prior model. Nvidia also claims the model is now 10 times faster overall than when DLSS 5 was first revealed in March.

Here is the part that matters for anyone reading a headline and expecting double the frame rate in their favorite game. The 2x figure describes the AI model's inference speed, not a guaranteed doubling of in-game performance. Nvidia's own game-specific number for NBA 2K27 is more modest: up to 1.4x higher frame rates. TechRadar, reporting on Nvidia's testing, framed the real-world NBA 2K27 gains closer to 30 to 40 percent. Those are solid numbers, but they are the difference between a faster engine and a faster car. As of this writing, no outlet has published independent frame-time benchmarks of the new model against the September version, so every public figure traces back to Nvidia's own release.

Nvidia doubled the speed of the AI model. It did not double the speed of the game. That gap is where marketing ends and benchmarks begin.

Where the new model lands first

Nvidia is staggering the rollout instead of flipping one global switch. NBA 2K27 gets the faster model on October 16, 2026, across GeForce RTX 50-series GPUs, RTX Spark laptops, and GeForce NOW. The new RTX Spark laptop line gets the model on day one of its October 16 launch, with Nvidia saying hundreds of games will be available at launch, including Fortnite, Cyberpunk 2077, and Marvel Rivals. RTX Remix, Nvidia's modding platform for remastering classic games, receives native DLSS 5 integration on October 23. Hogwarts Legacy follows on November 5, and Where Winds Meet is planned for late November.

The RTX Remix update is arguably the most interesting deployment for the enthusiast crowd. The October 23 release adds the accelerated 3D-Guided Neural Rendering model, a new high-speed Performance mode for frame generation, and developer tools that let modders mask scene objects, exclude individual elements from the neural network's view, and customize quality presets per 3D model. Nvidia also overhauled the AI texture generation workflow with ComfyUI Version 2 integration, replacing REST API node connections with a dedicated AI tools sidebar. Nvidia gave early builds to trusted creators: modders Sparkles and Kim2091 tested the stack on the in-development Fallout 4 RTX project, and enthusiast SoftSoundd benchmarked it in Mirror's Edge RTX mods, reporting improved reflections, contact shadows, and ambient occlusion accuracy.

Who gets it and who does not

The confirmed hardware list is narrower than many gamers would like. Nvidia names GeForce RTX 50-series desktop and laptop GPUs, the new RTX Spark laptop platform, and GeForce NOW's cloud streaming hardware. There is no mention of RTX 40-series, RTX 30-series, or older cards in the announcement. That fits Nvidia's established pattern: DLSS 3's frame generation launched RTX 40-series only, and DLSS 4's multi-frame generation launched RTX 50-series only. RTX 40-series owners, many of whom bought flagship cards within the last three years, are watching newer software features bypass them again. Nvidia has not addressed whether that will change.

The contrast with rivals is sharp. AMD's FSR and Intel's XeSS have generally prioritized broad compatibility across GPU generations, even at the cost of slower iteration. Nvidia's DLSS 5 update goes the other direction: tie the newest, most efficient model to the newest hardware. Which strategy earns more goodwill may depend less on benchmark numbers and more on how owners of two- or three-year-old cards feel about being excluded from each vendor's latest trick.

Why the VRAM cut matters right now

A 20 percent reduction in 4K VRAM usage would be welcome in any market. In this one, it reads as strategy. The RTX 5090 has been hard to find at retail through 2026, with Nvidia widely reported to be reallocating silicon toward AI data center and professional parts, and memory prices climbing across the board. A more efficient DLSS model stretches the usefulness of cards further down the stack: a 12GB or 16GB card that was tight on headroom at 4K suddenly has room to breathe, no hardware swap required.

That is convenient timing for Nvidia. A software update that makes mid-range and laptop GPUs look more capable costs the company almost nothing while its highest-end dies are busy elsewhere, and it softens the sting of a shortage Nvidia largely created by prioritizing AI silicon. It also gives Nvidia room to keep VRAM allocations conservative on upcoming SKUs: if a rumored 24GB RTX 5080 variant appears, the company can point to DLSS 5's reduced VRAM footprint to pitch it as flagship-adjacent even without the 5090's raw specs. Efficiency gains are genuine, but they also give a marketing team a lot of cover.

The honest read

The October DLSS 5 update is a legitimate technical achievement: a 24 percent smaller neural model that runs twice as fast with half the overhead and measurably less memory demand is the kind of compounding efficiency gain the industry rarely ships in one release. Gamers on RTX 50-series cards, RTX Spark laptops, and GeForce NOW get something real starting October 16, with NBA 2K27 as the first proof point.

The caveats are equally real. Every number traces to Nvidia. The headline 2x is a model metric, not a game metric. Older-card owners are excluded without explanation. And the VRAM savings land at exactly the moment Nvidia needs a software story to offset a hardware shortage. Treat the update as a genuinely better AI model wearing a carefully tailored business narrative, and wait for independent benchmarks before treating any of the claims as settled.