[CES 2025] Here’s All The New RTX Features NVIDIA Announced Alongside The RTX 50 GPUs

This is part 2 of the NVIDIA GPU launch coverage detailing all the software side of things for the new Blackwell-powered RTX 50 series GPUs. In case you aren’t familiar with that, click here for the rundown. Here’s an overview of all the new features the GPU maker has introduced, which will be available to all other RTX GPUs as well.

DLSS 4 (Multi Frame Generation)

The biggest talking point for the RTX 50 series is perhaps the DLSS 4, specifically the one feature in the entire graphics-upscaling feature suite: Multi Frame Generation, of MFG for short. This is the next step-up from RTX 40’s Frame Generation under DLSS 3, and it’s fairly self-explanatory. MFG essentially inserts up to three interpolated frames in between natively-rendered frames (which is pre-upscaled via DLSS Super Resolution), so theoretically you have 4x the output frame rate compared to native performance.

This is also the feature NVIDIA is latching on in their marketing promise of 2x performance over previous-gen counterparts, but there’s two major caveats. One, MFG only works on RTX 50 GPUs (in the keynote, Jensen Huang stated that the super-fast GDDR7 VRAM is one of the factors that makes MFG possible); and two, the game developers need to actively implement the feature into their games.

Oh, and there’s the third caveat: since this produces even more generated frames which is designed for you to perceive smoother motion, this doesn’t improve latency as the standard rendering pipeline is simply not involved most of the time. As usual, enabling FG or MFG will require NVIDIA Reflex to be activated at the same time to keep latency reasonable.

To enable MFG, you need to head to the NVIDIA App, under “Graphics”, select the game you wish to enable MFG and turn it on under the “DLSS Override – Frame Generation” option under the Driver Settings section. Here, you can select 3x or 4x mode depending on how many generated frames you wish to enable (standard FG is 2x, for reference – which is enabled via in-game settings). NVIDIA says 75 games will support DLSS 4 (MFG) from the get go, and it’s likely that all DLSS 3 titles are expected to support DLSS 4 at some point.

DLSS Updates For Existing RTX GPUs

For the rest of the RTX GPU owners, NVIDIA is also introducing upgrades to existing DLSS Super Resolution, Deep Learning Anti-Aliasing (DLAA), DLSS Ray Reconstruction, and DLSS Frame Generation. For DLSS Frame Generation, NVIDIA has introduced an enhanced Single-Frame Generation model that runs similar to MFG in RTX 50 GPUs. This results in slightly reduced VRAM usage and a small performance boost.

For Super Resolution, DLAA, and Ray Reconstruction, these upgrades involve the change from old convolutional neural network (CNN) model to the new Transformer model – the same model powering AI chatbots today – which results in sharper upscaling and reduced motion artifacts. All features will be available to existing RTX GPUs, with some features are currently under beta testing.

NVIDIA Reflex 2 (Frame Warp)

For competitive gaming and esports, NVIDIA says its new Reflex 2 latency-reduction feature has found new ways to cut latency even further. The key feature in the new Reflex iteration is called Frame Warp, which tells the in-game camera to move as soon as mouse input is detected, while ignoring the rendering process. This meant latency in THE FINALS is cut down to just 14 milliseconds, while VALORANT’s PC latency is shredded almost entirely to just one millisecond.

While this is indeed faster (NVIDIA’s data says 50% less latency over Reflex 1 and 75% less than native), it also creates “holes” (the white parts in the edge of the player’s weapon and screen) in the displayed frame that wasn’t rendered in time. To solve this artifact, the chipmaker uses a predictive rendering technique called “inpainting” to fill those gaps, relying on temporal and color data provided by the game engine. While the demonstration video shows pretty convincing results, it’s hard to say what happens under extremely fast mouse movements as it is not directly shown here.

NVIDIA says the new Reflex 2 feature will be first available to RTX 50 series GPUs on THE FINALS and VALORANT, with support for the remaining RTX GPUs coming “in a future update.”

Project G-Assist

The chipmaker has also provided a quick update on its Project G-Assist system assistant. This small-language model (SLM), Llama 3B-based chatbot is powered entirely onboard by any RTX GPUs, with only less than 1% of size of today’s LLMs – which will be designed to function as a personal assistant similar to Siri or Google Assistant in smartphones.

Some of the functions it can do includes adjusting performance, overclocking, generate performance metrics, and diagnostics; the chatbot also works with third-party peripherals from Logitech G, Corsair, MSI, and Nanoleaf. When active, the RTX GPU will assign some processing resources to the chatbot generate responses, and this can result in brief performance drops in games or GPU-heavy workloads. That’s normal, according to NVIDIA, and it’ll usually last a few seconds for each response.

That’s not all: NVIDIA also announces that it will publish a GitHub repository for users to add additional functionalities. For the general public, the Project G-Assist will be available in February as an optional component which can be downloaded via the NVIDIA App.

Pokdepinion: Good to see older RTX cards are still getting feature updates.