Beyond render: Nvidia brings a total real-time paradigm shift

Market analysis / Editorial review By GBAMFS Senior Market Contributor Martin Strnad

It is no exaggeration to state that we are at a pivotal moment in PC gaming history. Nvidia gave us a taste of what is to come with their announcement of the RTX 5000 series at CES 2025. Unveiled are next-generation GPUs featuring DLSS 4 multi-frame generation where old-school rasterization will be used to render only one out of every four frames the user sees on the screen, with the other 3 frames being AI-generated. The adoption of an AI graphics model in games is a shift of enormous magnitude, pointing to the inevitable sun-setting of rasterized 3D graphics.

How can we be sure that this shift is underway? Let us begin with the fact Nvidia is leagues ahead of their competitors at the present time, with the distance only growing between them in the area of four nanometer-on-down design. Next, we may examine the relative economies of producing ever-more-powerful GPUs that offer their customers only modest incremental performance gains–seldomer enticing many to make substantial investments in upgrades. Finally, it simply is too expensive for GPU makers to continue to maintain bleeding-edge R&D cycles with their production-lines being refreshed every two years.

This business model relies more on software updates containing better AI models that can boost performance, even those on the tensor cores of the past 3 generations of Nvidia GPUs. Where the main point of upgrading your GPU used to be getting more rasterization performance, we're now reaching the stage where the biggest reason to upgrade is for software updates which are usually artificially limited to the latest generation of GPUs (with the most recent tensor cores able to process more tokens than the previous ones.)

We're given a clear hint of Nvidia's aims for the future where the traditional method of rendering graphics will be just another guiding-input data point for their AI model. In the near-term, we can see that Nvidia leaving only 25% frames rendered on the rasterization method (with even that 25% heavily boosted by AI in the form of resolution upscaling,) paves the way for that outcome. Combining data from user input and visual styles, defined by training data of the given AI model, we will see these to be ultimately routinely used to generate the finalized graphics on-top.

I believe that once this paradigm shift fully kicks in, we will start seeing completely realistic-looking game content, as nobody can really copyright the dataset called "reality." And anything with styled visuals will rely on ability of game developers to create their own unique datasets which they then train their own custom AI models on. Considering this, we can clearly see why Nintendo (for example) so fiercely safeguards their visual assets—indispensable resources for their near future.

As a gamedev myself viewing these developments, I am led to believe that the future of art, writing, sound & music departments in game studios will completely change over to generating datasets for creating custom AI models, that will in turn cooperate as AI agents on end-user hardware. This will provide a distinct gaming experience for each user with its open-endedness and originality reliant on the teams' ability to create unique seeds and prompt-flows. Rather than relying on hard-coded scripting, game mechanics, and narrative sequences, as was done in the past.

This will prove to be a massive disruption of the software tools and workflows of big players within the industry. Epic, makers of Unreal Engine, and Unity are just dipping their toes into AI technology, principally for easing the burden of tedious tasks on their design departments. We have seen generative placement of visual assets in a scene in Unreal 5.x already, as well as using a small piece of environment sampled to be used to generate more like it at a larger scale. We have also seen experimental Indie games, alongside Skyrim and Fallout 4 mods using LLMs in cooperation with AI based voice generation to have much more lively conversations with NPCs.

What we are missing in this puzzle is introduction of AI prompting into the visual scripting editors of said SDKs as well as widespread adoption of gaming hardware that can run those models at acceptable speeds. That is precisely where the challenge remains. How much can be done by the AI and how much of the game has to be established beforehand by the game development team? And how do you even perform quality assurance on games with heavy use of generative AI models? Answering these questions is likely impossible without knowing the direction the early adopters of this technology like indie games will take it. I honestly do believe we will find out much sooner, though, than most think. “I believe that once this paradigm shift fully kicks in, we will start seeing completely realistic-looking game content”

There is an unresolved question: Will Nvidia be able to pull it off alone or will their competitors eventually jump on the same track, given Nvidia's software moat is already so difficult to cross? Another is will users actually accept this? I believe the answer is in the content itself. Like any new generation of hardware, it is paramount to have enough content running on it at a quality the customer base finds satisfactory.

As of now, the biggest selling point is the higher framerate being accelerated by AI technology; the price the end-user pays for this is a slightly higher input lag which ultimately matters only to the high-end users who care for the correspondence between framerate and their reaction time in games. While the largest cross-section of gamers mainly care if the game runs smoothly enough and just "looks good." It would make sense to expect some Nvidia-funded game projects will take a full advantage of this technology by completely hiding the rasterized visuals under fully AI-generated frames. If they are successful, they'll take the breath away from average gamers by achieving totally visually real-looking games. Even the animations of characters won't give it away, as those can be done via AI models as well.

On the horizon, if everything ends up working out in Nvidia's favor, we may actually see the return of Nvidia chips in gaming consoles and handhelds. With Nintendo Switch 2 likely to be the first and later with AMD potentially losing their dominance in the gaming hardware segment, a position they've held since PlayStation 4 and Xbox One. What is more is that the Switch 2 may prove to be the best way to showcase the technology as Nintendo's user base tends to not be super tech-savvy and doesn't really care about the intricacies of how graphics get rendered or how many extra milliseconds are added to their input lag. The patents Nintendo filed for their new hardware seem to be definitely pointing towards their heavy use of Nvidia's DLSS AI technology in Switch 2 even before any official announcements have been made. This could certainly give them a huge edge over any AMD or Intel APU-based PC handhelds in terms of visual fidelity on their new hardware.

It may come down to bit of a struggle with two parallel approaches: One where Nvidia pushes rasterization underneath AI technology and the other with AMD and Intel, unable to catch up to Nvidia's head-start, pushing towards more performance in rasterization at the cost of lower margins on their products. All in an attempt to appeal to more hardcore tech-savvy users. A struggle would ensue for developers and content creators who will have to make a decision as to where they will they invest a majority of their development time, with their choice coming down to how well each company can saturate the gaming hardware market, and with devs going where the most users are. Martin Strnad is a game design, narrative and production consultant. A veteran of 16 years in game development, Martin has also served as editor and contributor to European game news sites. On LinkedIn: https://www.linkedin.com/in/martin-strnad-15309338/ Copyright GBAMFS 2025. All Rights Reserved.