Double Creative Performance in 3D, Video, and Generative AI with New GeForce RTX-50 Series GPUs.
GeForce RTX-50 Series GPUs
New AI technologies improve livestreaming, DLSS 4 enhances 3D rendering, while NVIDIA NIM microservices and Blueprints supercharge AI on PCs. The GeForce RTX-50 Series was unveiled at CES and includes FP4 for advanced AI performance and up to three encoders with support for the 4:2:2 color scheme.
With new tools and capabilities for video, livestreaming, 3D, and more, the GeForce RTX 50 Series desktop and laptop GPUs, which were revealed today at the CES trade show, are set to drive the next wave of generative and agentic AI content production.
GeForce RTX-50 Series GPUs, which are based on the NVIDIA Blackwell architecture, can execute innovative generative AI models up to two times quicker and use less memory than their predecessors. They have up to 32GB of VRAM to handle large 3D projects, as well as ninth-generation NVIDIA encoders for sophisticated video editing and livestreaming.
RTX Video and RTX Remix upgrades, two new AI-powered NVIDIA Broadcast effects, and NVIDIA NIM microservices preconfigured and optimized models designed to accelerate AI content creation workflows on RTX AI PCs are just a few of the software improvements that come with these GPUs.
Built for the Generative AI Era
For creators, generative AI can produce amazing results, but even with the newest hardware, generative AI can be challenging to use as models become more complicated and larger.
To aid with this, the GeForce RTX 50 Series now supports FP4. Like file compression, FP4 is a lower quantization technique that reduces model sizes. GeForce RTX-50 Series GPUs offer more than two times the performance of the previous generation, and FP4 utilizes less than half the memory when compared to FP16, the default approach that most models use. With the help of NVIDIA TensorRT Model Optimizer’s sophisticated quantization techniques, this can be accomplished with almost minimal quality loss.
For instance, only professional GPUs and the GeForce RTX 4090 can handle Black Forest Labs’ FLUX.1 [dev] model at FP16, which requires more than 23GB of VRAM. FLUX.1 [dev] can run locally on more GeForce RTX GPUs with FP4 because it uses less than 10GB.
The FLUX.1 [dev] model can produce photos in 15 seconds with 30 steps using a GeForce RTX 4090 with FP16. Images can be produced in a little more than five seconds using a GeForce RTX 5090 with FP4.
Artists will have more control over text-based image production with a new NVIDIA AI Blueprint for 3D-guided generative AI based on FLUX.1 [dev], which will be available as an NVIDIA NIM microservice. This blueprint enables designers to control AI image development by using basic 3D objects, either manually built or generated by AI, and arranging them in a 3D renderer such as Blender.
High-quality photos that complement the composition of the 3D scene may subsequently be produced using a prepackaged workflow driven by ComfyUI and the FLUX NIM microservice.
In February, a one-click installer for the NVIDIA Blueprint for 3D-guided generative AI is anticipated to be made accessible on GitHub.
This month, Stability AI’s Stable Point Aware 3D, or SPAR3D, model will be accessible on RTX AI PCs, the company revealed. By enabling real-time editing and the capacity to create an item from a single image in less than a second, Stability AI’s new model will revolutionize 3D design and provide remarkable control over the creation of 3D content. This is made possible by RTX acceleration.
Professional-Grade Video for All
With support for the 4:2:2 pro-grade colour standard, Multiview-HEVC (MV-HEVC) for 3D and virtual reality (VR) video, and the new AV1 Ultra High Quality mode, GeForce RTX-50 Series GPUs offer a generational leap in NVIDIA encoders and decoders.
Color information is reduced by the 4:2:0 color compression used by the majority of consumer cameras. Generally speaking, 4:2:0 is enough for video viewing in browsers, but it lacks the color depth required for sophisticated video editors to color grade videos. With only a 1.3x increase in RAW file size, the 4:2:2 format offers twice as much color information, making it the perfect balance for video editing operations.
Because 4:2:2 videos have larger file sizes, decoding them can be difficult. Multi-camera video editing with up to eight 4K 60 fps inputs per decoder is possible with GeForce RTX-50 Series GPUs.
GeForce RTX 5090 has three encoders and two decoders, while 5080, 5070 Ti, and 5070 have two. Multi-encoder and decoder combinations and faster GPUs make the GeForce RTX 5090 output video 60% faster than the 4090 and four times faster than the 3090.
GeForce RTX-50 Series GPUs can also use the ninth-generation NVIDIA video encoder, NVENC, which improves HEVC and AV1 encoding (BD-BR) video quality by 5%, and a new AV1 Ultra Quality option that compresses 5% more at the same quality. Also included is the sixth-generation NVIDIA decoder, which doubles H.264 video decoding performance.
In February, NVIDIA will integrate these technologies with Adobe Premiere Pro, Wondershare Filmora, Capcut, and DaVinci Resolve from Blackmagic Design.
