Alongside the launch of the new GeForce RTX 50 series GPUs, CES 2025 also saw the debut of the NVIDIA RTX Kit, a new suite of neural rendering technologies that promises to help game developers make the next leap in graphics.
The NVIDIA RTX Kit covers a wide range of uses, from AI-accelerated ray tracing to more lifelike digital humans and the rendering of scenes with massive amounts of geometry.
Let's start with the most notable innovation. With the Blackwell architecture, NVIDIA is introducing RTX Neural Shaders. These are small neural networks designed to be built into programmable shaders to deliver a variety of useful effects. With the upcoming RTX Neural Shaders SDK, game developers can train game data and shader code on an RTX AI PC accelerated by NVIDIA Tensor Cores. During training, the neural game data is compared against the output of traditional data and refined over many iterations.
For now, RTX Neural Shaders have three main applications, all of which we have covered in NVIDIA's research papers: RTX Neural Texture Compression, RTX Neural Materials and RTX Neural Radiance Cache. RTX Neural Texture Compression is valuable because it can compress thousands of textures in under a minute while saving up to 7 times the VRAM compared with traditional block-compressed textures, with no negative impact on quality. In the new tech demo, the RTX NTC textures take up just 80MB, while conventional block-compressed textures need as much as 520MB of VRAM.
RTX Neural Materials uses artificial intelligence to compress complex shader code that is normally reserved for offline rendering, making material processing up to five times faster and allowing higher-quality assets to be available in games.
Finally, artificial intelligence is applied in RTX Neural Radiance Cache to infer countless light bounces after the first one or two, which greatly improves indirect lighting and performance. NRC has been around for a while as part of the RTX GI SDK, and it will soon be introduced in RTX Remix and Portal with RTX.
The RTX Neural Shaders and RTX Neural Texture Compression SDKs will be added to the NVIDIA RTX Kit later this month. Today, Microsoft also announced that DirectX will soon support Cooperative Vectors, delivering higher performance on GeForce RTX GPUs when using RTX Neural Shaders. Bryan Langley, a program manager on Microsoft's partner team, said:
"Microsoft is excited to work with NVIDIA to advance the next generation of graphics programming by bringing neural shading technology support across the industry. DirectX will soon support Cooperative Vectors, unlocking the power of Tensor Cores on NVIDIA GeForce RTX hardware and helping game developers fully accelerate neural shaders on Windows."
NVIDIA's RTX Kit also includes RTX Mega Geometry technology, which will debut in Remedy's Alan Wake 2 with a free update later this year. A demo video at CES 2025 showed off the technology, which is meant to improve ray tracing in scenes with a lot of complex geometry, like in games that use Unreal Engine 5's Nanite. According to NVIDIA, RTX Mega Geometry speeds up BVH building, allowing up to a hundred times more triangles to be ray traced than today's standard. The technology will be released later this month in the RTX Kit, and will soon be available through NVIDIA's RTX branch of Unreal Engine.
RTX Neural Faces is a tool in the NVIDIA RTX Kit that takes a rasterized face image along with 3D position data and generates a more natural-looking face through a real-time generative AI model. The generated face is trained on thousands of images created offline, covering every angle and a range of lighting conditions, emotions and occlusions. The training process can use real-world images or AI-generated ones. The model is then optimized with NVIDIA TensorRT to infer faces in real time.
This works together with the RTX Character Rendering SDK to produce hair and skin. With the new GeForce RTX 50 Series processors, NVIDIA has added linear swept spheres (LSS), a GPU acceleration method that reduces the amount of geometry needed to render strands of hair. LSS uses spheres instead of triangles to match the shape of hair more accurately. LSS lets ray-traced hair run with better performance and take up less memory. The Skin and Hair algorithms that use the RTX Character Rendering SDK will launch later this month, along with the NVIDIA RTX Kit. The hair algorithm will appear in Indiana Jones and the Great Circle with an upcoming free update. The skin algorithm will also be available through RTX Remix.
Last but not least, NVIDIA has promised a major update to its Audio2Face technology. Frankly, we weren't really impressed with how it was implemented in GSC Game World's STALKER 2; the AI fell short at producing good facial movements from audio input. The new short demo video, however, looks considerably more expressive. According to NVIDIA, a new real-time diffusion architecture for Audio2Face has been deployed to deliver more accurate and detailed lip sync, lively emotion, and natural nonverbal reactions.
https://www.youtube.com/watch?v=dm8-gNin76c
Source: WCCFTECH
