Yesterday, Nvidia hosted a Central European media briefing where they showcased its latest technological developments live. Unsurprisingly, the spotlight was on innovations unveiled or launched last month at CES 2026—namely the G-Sync Pulsar technology for monitors and DLSS 4.5, both of which we have already covered in separate articles—as well as a demo of the ACE technology, which aims to bring AI-driven assistants into games.
In its demos, Nvidia heavily promoted DLSS 4.5 upscaling and its associated frame generation, including a new technology capable of inserting up to five AI-interpolated frames after each native frame, along with dynamic frame generation that allows the number of AI-generated frames to be switched during gameplay. However, this is difficult to convey in text, so we will focus on the other showcased technologies.
How to imagine the effect of G-Sync Pulsar
The problem with monitors is that you need to see them in person. This is especially true for any backlight strobing technology such as he freshly launched G-Sync Pulsar—which can now coexist with adaptive refresh, unlike Nvidia’s previous ULMB and enhanced ULMB 2, which were not compatible.
Below you can see an animated capture from a demo video Nvidia uses to demonstrate the improvement in perceived motion clarity enabled by backlight strobing (in this case, Pulsar). But be aware that this is not a real camera recording showing how two actual monitors behave.
It is an attempt to visualize the effect of the technology through a constructed animation, since the resulting perception is inherently fleeting and difficult to capture. This is a general issue in all marketing related to monitor response times, where manufacturers are forced to use artificial sharp and blurred images to explain motion blur reduction—yet those images are not actually capturing any reality.
According to Nvidia, the video is meant to convey that G-Sync Pulsar subjectively improves motion clarity by roughly 4×—that is, a fourfold reduction in perceived motion blur—based on internal evaluation methods. The demo aims to approximate this, but we cannot guarantee it is fully representative.
The video is slowed down and artificially created to remotely convey this subjective perception; it may not be entirely accurate. Do not interpret it too literally. Notice that in the lower video representing Pulsar, you do not see the flickering due to backlight strobing (on-off switching, or more precisely in the Pulsar’s case, a black band periodically rolling across the screen) that would normally become visible when slowed down. Instead, it shows consecutive frames with reduced blur—which Pulsar does not literally do, but this is how the eye should ideally perceive the result.
The strobing backlight should work on the brain by making you consciously perceive only the illuminated image, while the periods when the monitor is dark go unnoticed. This would not be visible in a real slow-motion camera recording, although such footage could at least show the blur reduction capability within individual frames.
In the upper video (without Pulsar), only blurred frames representing response artifacts are visualized. If you slowed down the image, the blur would likely also cycle, as pixels initially start in the incorrect state from the previous frame and gradually transition to the correct colors before the next refresh occurs. Note that the frames don’t actually remain fully blurred for their entire duration; otherwise, turning off the backlight during the initial portion of the frame via Pulsar would not be able to improve the image seen.
How does the strobing backlight work?
That said, we are not suggesting you should dismiss this demo. The principle that strobingbacklighting affects human visual perception in this way is well known and considered a documented fact—and it was indeed visible in the demos. There are two reasons for this. The first is easy to understand—the image is dimmed just before refresh and remains dark for a portion at the beginning of each frame, hiding response artifacts caused by the relatively slow switching of LCD pixels (as well as artifacts from aggressive overdrive). Once the backlight turns on, the worst of the response time artifacts should ideally be gone or significantly reduced.
Your vision then registers this cleaner image when it appears instead of noticing and “holding in the brain” the imperfect initial state of the frame. This effect is particularly noticeable when scrolling a block of text on the screen, where response artifacts heavily manifest as perceived motion blur—something most users have experienced while browsing the web on laptops or LCD-based PC monitors. With Pulsar-enabled monitors, scrolling text appears significantly clearer and less blurred (although whether you would enable this mode outside gaming is another question).
There is also a second factor: When viewing LCD or OLED monitor screen that shows image uninterruptedly, simply switching frames at the time or refreshes, human vision ironically struggles to perceive the images as smooth motion. The brain subconsciously registers that the image is static for periods and then “jumps” forward when the frame changes. The continuous display of the image ironically reduces the brain’s ability to fall for the illusion and perceive the series of images as a smooth motion—making the image appear blurrier than it objectively is on an LCD panel.
Conversely, it’s a paradox but the brain is more willing to “stitch together” into smooth motion illusion such frames that are shown on a flickering displays, and the result then subjectively appears less blurred. Film projectors exploit this effect using a rotating shutter. This is the second mechanism that reduces the perception of motion blur.
G-Sync Pulsar as an OLED alternative?
Unfortunately, this implies that even OLED displays—which have much faster response times and might not need pulsing for that reason only—may still suffer from the “psychological” effect related to the brain’s processing of motion. At present, we do not know whether Nvidia plans to implement Pulsar on OLED panels.
In that case, the pixels themselves would need to be turned off and on, meaning their color states would switch. If the image were to be strobed, we would still be dealing with the objective, physically real causes of motion blur, since OLED pixels do not have truly zero response time either. We asked whether Nvidia—or display manufacturers—are conducting studies or research in this area, and were told that internally, the Pulsar team would indeed like to see this (Pulsar with OLED) happen one day—which is encouraging.
However, because the strobing mechanism would have to be introduced differently than via traditional backlight switching, a viable method of implementing Pulsar on OLED is not yet perhaps established. Nvidia representatives did mention that engineers are exploring ways to improve motion perception on OLED, though without disclosing details—it may not necessarily involve making the display flicker.
Until then, G-Sync Pulsar can serve as a competitor to OLED by overcoming some of LCD’s disadvantages—without introducing OLED’s longevity and burn-in concerns (although lower contrast remains a drawback od LCD panels). The downside is that Pulsar requires advanced monitor electronics, meaning it will likely remain confined to relatively expensive specialized gaming monitors.
Nvidia ACE: The arrival of an AI assistant in games
Nvidia ACE (Avatar Cloud Engine) was first announced nearly three years ago, during Computex 2023. At the time, we wrote that the technology was intended to simulate autonomously thinking characters in games—although this naturally raises potential issues similar to those seen with chatbots, such as unpredictable output or the so-called AI hallucinations.
Nvidia has now presented another potential use case for ACE—as an in-game “AI assistant.” The demo showcased ACE integration in Total War: Pharaoh, where ACE powers the player’s advisor, offering strategic suggestions for various gameplay situations. Instead of predefined scripts and hints, the advisor’s output is generated by an AI model trained on, in addition to the game’s wiki, game data, rules, unit statistics and mechanics (as well as information available on the game’s online wiki, as another data source). The model also access to current in-game internal data and metrics to know what is happening in the game.
In the demonstration, a rebellion within the empire led to a loss of control over part of the territory. The advisor suggested ways to improve public satisfaction and prevent or resolve rebellions—for example, by constructing buildings that provide relevant bonuses. It also recommended which units to recruit, and so on. Essentially, it replaces a traditional help and hint system with an interface that simulates natural communication—similar to internet chatbots.
Whether you want to use such a feature or consider it necessary is up to you. As with the broader rise of chatbots in general, integration may sometimes feel unnecessary and driven primarily by AI being the fad du jour, while in other cases it could be a good addition and purposeful part of gameplay.
The system does not use a large language model (LLM) like ChatGPT for example, as those run in the cloud on powerful GPU accelerators, but rather an SLM—a small language model. It cannot be as capable or versatile (though it is more narrowly specialized), but this is necessary for the model to run directly on the PC’s graphics card. In other words, inference is performed locally—no requests are sent to a server. The model size must be limited not only to fit within GPU memory—which is not exactly abundant these days—but also because it adds to the game’s storage requirements on the PC’s SSD.
This implementation of ACE in Total War: Pharaoh is expected to become publicly available this year—specifically entering a closed beta phase, meaning not fully public, but accessible beyond Nvidia’s internal testing.
According to information from the demo, the technology should not be exclusively limited to Nvidia GPUs, though performance of such integrated chatbot on competing GPUs may be lower. The details of implementation and GPU compatibility will ultimately be determined by individual game developers.
Sources: Nvidia, on-site reporting
English translation and edit by Jozef Dudáš
⠀
