Path tracing has become one of the clearest showcases of modern PC graphics, but the visual cost is still obvious. Even with DLSS 4.5 and its Ray Reconstruction push, game engines are not yet free from the artifacts that appear when image quality and performance are forced to share the same budget.
The core issue is not a lack of ambition. Nvidia is preparing a second-generation Ray Reconstruction update for August, yet the bigger question is whether that step can finally make path tracing feel complete without introducing new compromises elsewhere.
Why the latest AI tools still matter
DLSS 4.5 has been positioned as an important part of the modern graphics stack, especially when paired with frame generation and path tracing. Its transformer-based model has already improved Nvidia’s ability to handle visual artifacts inside a frame, but that does not mean the underlying problems have disappeared.
In busy scenes, the weaknesses are still easy to spot. Dense NPC areas, fast-moving objects, and complex foliage can trigger temporal and spatial inconsistencies that leave the image softer, blurrier in motion, or less detailed than expected.
That is where Ray Reconstruction enters the picture. The feature is designed to replace traditional lightweight denoisers with a neural network trained on supercomputers, helping to cover the gaps created when path tracing runs with very low ray sample counts to stay playable.
The remaining visual trade-offs
Modern reconstruction techniques still sit on top of the same longstanding temporal anti-aliasing foundation. TAA is efficient for static scenes such as walls and signs, but it struggles when the camera moves quickly or the screen is crowded with particles and active objects.
That weakness continues to show up as ghosting, motion smearing, and the loss of fine detail. In some games, the image can even look worse without an upscaler, which shows how dependent current rendering pipelines have become on temporal methods to hold visual quality together.
Crimson Desert has been highlighted as one example of the challenge. The game uses heavy path tracing, and when Ray Reconstruction 1.0 was enabled to address the issue, it introduced new problems such as visible noise and “boiling” artifacts on reflective surfaces.
Later updates reportedly cleaned up some of those shortcomings, but the technology is still not considered fully solved in real-world use. That is why Nvidia’s upcoming second-generation Ray Reconstruction is drawing attention now.
What DLSS 4.5 is supposed to improve
According to Nvidia, the second-generation transformer denoiser is meant to reduce aggressive ghosting, trailing on moving characters, persistent indoor noise, and the rougher volumetric clarity issues that appear around particle effects. If those claims hold up in practice, the update could close some of the most visible gaps left by traditional rendering paths.
The catch is that better image cleanup comes with a cost. Ray Reconstruction runs in the same pipeline as DLSS and relies heavily on Tensor Cores, which adds GPU workload and can reduce performance in a noticeable way.
Nvidia has also said the second-generation transformer model in DLSS 4.5 processes 20% more parameters and delivers 35% more compute capability. That points to a more capable AI model per scene, but it does not yet answer whether the overhead has been reduced enough to keep the feature efficient.
Hardware support remains a dividing line
The performance burden is especially relevant on older hardware. The latest transformer model requires hardware-accelerated FP8 instruction support, which is natively available on Ada Lovelace and Blackwell.
Without native FP8 support, the compute overhead can hit performance much harder. That leaves RTX 30 series cards and older GPUs at risk of paying a larger penalty when trying to run the DLSS 4.5 transformer model.
Frame generation is often used to offset those demands and push games toward playable 60 FPS at high resolutions. But higher frame rates do not automatically remove the temporal artifacts these systems were built to reduce, and in some cases frame generation can preserve or even amplify ghosting and motion smearing.
That means the number on the frame counter may improve while motion quality stays uneven. For now, path tracing, DLSS, Ray Reconstruction, and frame generation still work more like layered compromises than a final solution to the visual problems of modern PC gaming.
The practical value of Ray Reconstruction remains clear, especially in heavy path tracing scenarios where low-sample rendering needs help. August will show whether the second-generation update is a meaningful leap forward, or only another refinement in a graphics pipeline that still asks players to trade visual purity for performance.