NVIDIA DLSS 5 Brings Generative AI Into Game Rendering

DLSS 5 is emerging as a new approach to game rendering, using generative AI to change how characters, lighting, shadows, geometry, and other visual elements appear on screen. Early testing suggests the technology can produce changes beyond traditional upscaling, though the performance cost remains a major consideration.

As graphics technology continues to develop, neural rendering is becoming part of the conversation around the next generation of game visuals. DLSS 5 uses algorithms trained to apply generative AI to different elements of a game, including lighting, subsurface scattering, shadows, geometry, and character details.

DLSS 5 Brings Generative AI Into Game Rendering

We have been testing DLSS 5 across many games, including Final Fantasy VII Rebirth, with Final Fantasy XVI and other titles also planned. The results show the technology can change character appearance even when the underlying game settings are reduced. One interesting part is that DLSS 5 can still improve characters even when the game is set to low settings.

It doesn't simply replace characters with something unrelated. The characters remain representative of their original designs while receiving changes to their visual presentation. The technology also includes sliders that let you adjust the effect. For much of the testing, we pushed the settings all the way up because that makes the difference easier to see.

Final Fantasy VII Rebirth provided one of the clearest examples during testing. Cloud already has a specific visual style, and DLSS 5 changes the way his features are presented. When we turned on DLSS 5, the difference was immediately noticeable. The technology amplifies features that are already present in the characters instead of completely changing their designs.

The same applies to characters such as Tifa and Aerith. Switching between the original rendering and DLSS 5 shows how the technology changes lighting, skin tones, and other details. These changes can bring out details that traditional rendering does not emphasize in the same way.

Neural Rendering Could Change More Than Characters

Current testing focuses heavily on characters, but DLSS 5 is also expected to apply neural rendering to other parts of a game. Vegetation, world geometry, ground textures, and other elements could receive similar treatment. That means the impact could extend beyond character models and affect how entire game environments are presented.

We have seen major changes in game graphics over the years, but this approach represents a different direction. Instead of relying only on traditional rendering techniques and more powerful hardware, generative AI can alter the final visual result. The result is a rendering pipeline where AI becomes part of how the game image is produced, rather than simply upscaling a traditionally rendered frame.

The biggest concern from the current testing is performance. We are using an RTX 5090, which is a $5,000 graphics card in the current setup. Even at that level, enabling DLSS 5 can cut performance in half. That means frame generation becomes necessary to bring the frame rate back to a smooth and playable level.

The final experience can still be smooth, but the performance cost is difficult to ignore. The question gets more complicated when you consider other GPUs. What happens on an RTX 5070 Ti, RTX 5080, RTX 5060 Ti, or RTX 5050? What about older cards such as the RTX 4090, RTX 4080, or RTX 4070? If a $5,000 GPU is already taking a major performance hit with the current version of DLSS 5, lower-end hardware could face even bigger limitations.

AMD Does Not Need to Panic Yet

Current performance requirements are one reason AMD doesn't necessarily need to worry about DLSS 5 right now. The version being tested is not the final official implementation. When the technology is released, performance could improve. NVIDIA will also likely keep optimizing the technology before it reaches users in its final form.

When NVIDIA first demonstrated DLSS 5, it used two RTX 5090 GPUs. One handled the game while the other handled the upscaling and neural processing. That setup showed how demanding the technology could be. The current implementation therefore may not reflect where performance ends up.

Even if AMD doesn't need to react immediately, the company needs to watch what NVIDIA is doing with neural rendering. AMD could potentially bring neural rendering to the RX 9070 XT. However, performance could be closer to what an RTX 5070 Ti or RTX 5070 delivers with the technology, depending on AMD's underlying silicon and how the implementation works.

RDNA 5 needs to answer this question with its feature set.

If AMD doesn't develop comparable neural rendering technology, the company could eventually fall behind in visual quality, even if its traditional rendering performance remains competitive. The more important question is what comes next.

Game graphics have evolved from 8-bit games to increasingly complex rendering techniques, and DLSS 5 represents another shift in that progression. Calling the technology unnecessary or dismissing it without seeing the results does not change what it can do. Once you compare the output directly, the difference becomes something to judge on its own merits.

The main limitation right now is performance. If NVIDIA can reduce that cost while maintaining the visual improvements, DLSS 5 could become an important part of future game rendering. For AMD, that means neural rendering cannot be ignored. RDNA 5 will need to address this direction if AMD wants to remain competitive as game graphics continue moving toward AI-assisted rendering.