NVIDIA GeForce Now And DLSS Get Glorious Upgrades At CES 2026
Enter G-Sync Pulsar Gaming Monitor Technology
The biggest announcement out of NVIDIA today, in our opinion, is the news that G-SYNC Pulsar is finally coming to market. We've been excited about Pulsar since we first heard about it way back at CES 2024, two years ago. If you're not familiar, this technology allows gamers to combine G-SYNC variable refresh rate smoothness with ULMB clear motion. While ASUS (with its ELMB Sync) and other vendors have released displays that can combine VRR and motion clarity strobing, NVIDIA's solution was built from the ground up for this purpose, and promises perfectly clear motion without overdrive foibles, or having to worry about staying in a monitor's "VRR range".
NVIDIA's adding an interesting new feature to G-SYNC Pulsar that wasn't previously disclosed; apparently all G-SYNC Pulsar monitors will include "Ambient Adaptive" technology that will auto-adjust brightness and colors based on your ambient lighting. This feature will probably be included with a switch to disable it; at least, we certainly hope so, because our experiences with this kind of technology aren't particularly flattering. But as they say, don't knock it until you try it.
The First Crop Of G-Sync Pulsar Monitors
The presentation presented four specific models of G-SYNC Pulsar display that it says will be on sale this Wednesday: Acer's Predator XB273U F5, AOC's AGON PRO AG276QSG2, ASUS' ROG Strix Pulsar XG27AQNGV, and MSI's MPG 272QRF X36. All four of these displays seem to be 27" monitors in QHD (2560×1440) resolution with maximum refresh rates of 360 Hz that will apparently start at $599. That's all fine and well, but where are the 4K HDR G-SYNC Pulsar displays, eh? Eh?
DLSS 4.5 Brings Dynamic MFG And Second-Gen Transformer Tech
It's strange days when a software update for your GPU can give you a phenomenal experiential improvement, but that's exactly what happened with the release of DLSS 4. The new transformer model for DLSS upscaling drastically improved many areas of the technology for RTX gamers from Turing to Blackwell, particularly the way it handles animated textures and fine detail. Now, NVIDIA's launching DLSS 4.5, with a new model for Super Resolution that it says makes for better temporal stability, reduced ghosting, and smoother edges.
We actually ran into this issue that NVIDIA is showing in our own testing of The Elder Scrolls Part IV Oblivion Remastered; even when using DLSS 4 Super Resolution, you can see very obvious visible ghosting as the player character waves a weapon around simply when moving. Apparently the new update resolves this, which is great. This is one of very few titles where we've observed this kind of ghosting, but it's great to see NVIDIA continuing to iterate on and improve its technology.
The other big GeForce technology that's getting upgrades is Multi Frame Generation. That's right, you can now use MFG up to 6x, so your 40 FPS game can look like it's running at 240 FPS while still feeling like a 40 FPS game! To be frank, we don't know why this arbitrary limit exists because other technologies have already proved that multi-frame generation as high as 20X is possible, even trivial, but it might be because the AI kernel that NVIDIA uses for MFG is too demanding. In any case, we're not wild about MFG with a fixed scalar.
However, we are pretty interested in the new Dynamic Multi-Frame Generation. In essence, you get your game running at a satisfactory frame rate using Super Resolution, and then you flick on Dynamic Multi-Frame Generation, and the GPU will generate extra frames to get you to your monitor's maximum refresh rate. While not novel at all, this is a very cool idea that, arguably, should have been the way gamers were initially introduced to frame generation to begin with. In theory, this method should produce perfect frame pacing and eliminate judder entirely. In theory.
These days NVIDIA can't let an announcement go without publishing a really controversial benchmark slide, it seems. This slide seems to be claiming that it is good and even preferable to play games that are struggling to hit even 30 FPS on a GeForce RTX 5090 at upwards of 400 FPS using frame generation. That's not actually what it's saying, though; remember that NVIDIA likes to fold in the performance upgrades from upscaling when talking about frame generation. All NVIDIA is really saying here is that DLSS makes path-traced games playable, which is well-known and arguably the reason the technology exists in the first place.
As before, you can enable DLSS 4.5 features in the free NVIDIA App, which doesn't require an online-enabled login like GeForce Experience used to, by the way. You can set DLSS FG to Dynamic mode and set a custom FPS target or, if you prefer, simply set your monitor's refresh rate as the target. This is really intriguing and we can't wait to test it once the update arrives. NVIDIA says the DLSS 4.5 Super Resolution upgrade is available now with the latest GeForce driver, while Dynamic Frame Generation is apparently coming in the spring, potentially only for Blackwell GPUs, although that's not quite clear from the materials we were given.
NVIDIA's Extra Gaming Goodies For GeForce PC Gamers
NVIDIA has a couple of other cool announcements related to gaming on GeForce GPUs. The first has to do with RTX Remix. In case you aren't aware, RTX Remix is a toolkit for game modders to create path-traced "RTX" versions of classic games. We've looked at a couple in fine detail, including Portal with RTX and Half-Life 2 RTX. It's not just Source-engine games that are getting upgrades, though. In fact, NVIDIA says there are over 120 RTX Remix mods in progress, for games like Vampire the Masquerade Bloodlines, Sonic Adventure, and Left 4 Dead 2.
Naturally, NVIDIA has some news regarding RTX Remix. The latest version of the SDK will soon support "RTX Remix Logic", which ties the Remix renderer into the game state of the game bidirectionally. This allows game events to directly affect the state of the RTX Remix graphics. Before, RTX Remix, despite being fully path-traced and dynamic in that way, was actually quite static in terms of how the renderer was defined for a given scene. Now, RTX Remix visual effects can actually change based on the game state and events in the game world.
Whatever you think about game streaming, most reviews agree that GeForce NOW is one of the less-objectionable services. Not only does it work well for what it is, but instead of buying games that are restricted to GeForce NOW (like Google's defunct Stadia), you are presented with a Windows desktop where you login to your own Steam account, your own GOG account, and so on. For $20/month, if the experience is satisfactory where you live, it's not a terrible idea. You'd have to spend over 8 years on the service to make up the price of a GeForce RTX 5080 gaming PC.
NVIDIA's announcements concerning GeForce NOW today are basically threefold: the company is adding native clients for both linux-based PCs as well as Amazon Fire TV devices, which are both good news. The latest update will have support for flight controls, making MSFS and X-Plane fans happy, and it's also adding Single Sign-On support for Microsoft-Activision-Blizzard's Battle.Net client as well as Gaijin Entertainment's launcher for games like War Thunder, Enlisted, and Crossout. That's a healthy quality-of-life upgrade for GeForce NOW users.
Naturally, NVIDIA's All About AI At CES 2026
We've saved the non-DLSS AI announcements for last, because of course you already know that NVIDIA is a big proponent of pushing AI into every space possible. The company highlighted two games that have or are integrating its ACE technology: The Sims-like life simulator title inZOI, which has AI-powered NPC citizens, and PlayerUnknown's Battlegrounds, which has AI-powered "allies" coming this year.
NVIDIA points out that "depth and complexity are hallmarks of PC gaming." It's definitely true that PCs have historically played host to more intellectually challenging experiences, like large-scale simulation, 4X, and grand strategy titles. So, what if you didn't have to hit the wiki every 45 seconds to understand what was happening in your games? That's basically the pitch for NVIDIA's in-game advisor with ACE.
Coming to Total War: Pharaoh this year in a playtest: AI-powered advisors that you can ask for help with your game. They'll be powered by the ACE SLM (Small Language Model). Now, SLMs are prone to hallucination, so to help get around this, the system uses a tightly constrained context space for the game and emphasizes heavy tool use, including Retrieval-Augmented Generation using the live game state and over 1200 interlinked tables it can reference for game mechanics data. It looks interesting, but this writer isn't yet convinced it is more useful or reliable than opening Chrome in on his second monitor.
Finally, NVIDIA talks about running generative AI workloads on home PCs, and this is where we have to pump the brakes just a little bit. The company points out that interest in local AI is growing with numbers from GitHub and HuggingFace, and that's definitely true. However, we have to take issue with the "AI PC Ecosystem Primed For Breakout 2026" headline; the memory shortage—largely fuelled by NVIDIA and its institutional customers—is likely to cause a massive downturn in the PC market this year, according to IDC. NVIDIA itself is even rumored to be slashing GeForce GPU production because of it.
This slide is also dubious. While we applaud NVIDIA's hard work in accelerating local AI, we've tested quantized models and the idea that they provide the same results as full-precision models is, let's be nice and say, optimistic. With that in mind, NVIDIA comparing ComfyUI performance before and after NVFP4 quantization and then crowing about a 4.6x speedup feels a bit like being told that your car gained 10 MPG fuel efficiency by ripping out the air conditioner and the radio. It's faster, sure, but it's not the same result, and comparing them feels a little disingenuous.
NVIDIA also proudly reports that you can reduce the memory usage of bleeding-edge image generation models FLUX.2 and Qwen Image down to just 26GB and 18GB respectively by quantizing them to NVFP4. Well, sure—but we feel compelled to point out that, besides the other effects of quantization that aren't covered here, those sizes are still too big for anything short of a GeForce RTX "x90" card. So if you have a $2,000 GPU, hurray! You can run quantized versions of high-end image generation models. Is it cool that you can do this locally? Yes, absolutely—but it's worth being clear about how slim the audience for what NVIDIA is boasting about here actually is.
To be fair, the company also reports that it has cut Black Forest Labs' older FLUX.1 model down to just 9GB, which absolutely will run on even a GeForce RTX 3060, and FLUX.1 is still very, very good. We haven't tested this quantized version, but we have tested an FP8 version on a competitor's hardware and found the results to still be quite impressive. So kudos to NVIDIA for its continuing research in this area; we just think presenting the data without the context of who it applies to (within the context of a "GeForce" presentation) is slightly self-aggrandizing.
Along similar lines, NVIDIA is announcing that the Lightricks LTX-2 audio-video generation model is now available for download, and optimized for GeForce RTX GPUs. For context, the original LTX-Video, while not necessarily the absolute best audio-video model available for local generation, is still very high quality and most importantly, can do its thing at real-time speeds if you have the appropriate hardware. Apparently, LTX-2 raises the maximum resolution to a full 4K, and allows clips to be up to 20 seconds in length, which is incredibly impressive. NVIDIA says LTX-2 is available now on HuggingFace.
Videos generated by LTX-2 can be upscaled through RTX Video Super Resolution, which undoubtedly improves quality a bit, although NVIDIA's slide above is once again blurring the truth a little bit. What's making the performance difference is the use of the quantized model in the generation stage, not so much the video upscaling solution. Still, we don't doubt that the NVIDIA-optimized pipeline is faster.
NVIDIA's CES 2026 Show Proves Team Green Still Cares About Gamers
NVIDIA is one of the most valuable companies in the world now, a position it reached largely on the back of untold demand for its cutting-edge AI accelerators. Critics often note (correctly) that gaming represents a tiny fraction of NVIDIA's revenue these days, but we don't believe it when they extrapolate to say that "NVIDIA will exit the gaming market." Doing so would be a huge middle finger to the face of the very people who made the company what it is.
We go deep on DLSS 4.5 And G-Sync Pulsar With Justin From NVIDIA
We could talk at length about the validity and stability of the AI boom, but the fact of the matter is that it doesn't matter. NVIDIA's origins are in PC gaming and graphics, and Jensen Huang has repeatedly reaffirmed his commitments to that market and to GeForce gamers. NVIDIA's actions back that up: the company has showed absolutely no signs of slowing when it comes to research and development for PC gaming. While we doubt that Jensen Huang has much time to spend on gaming himself these days, it's clear that the teams building GeForce still do, and that's ultimately what really matters.
NVIDIA's got other announcements this CES besides this GeForce group; if you're interested in the AI side of things, you should check out the company's news concerning the latest updates to its DGX Spark platform.
