From NVIDIA: 2025-01-08 09:00:37

Generative AI has revolutionized various aspects of life, from writing to gaming. PC enthusiasts are at the forefront of exploring this technology. NVIDIA’s RTX AI Garage series aims to educate developers and enthusiasts on NIM microservices and AI Blueprints for AI agents, digital humans, and more.

At CES, NVIDIA unveiled new AI foundation models on RTX AI PCs powered by GeForce RTX 50 Series GPUs. These GPUs deliver up to 3,352 trillion AI operations per second and feature FP4 compute for enhanced AI inference performance. NVIDIA AI Blueprints offer ready-to-use workflows for digital humans and content creation.

NVIDIA NIM provides prepackaged AI models optimized for PCs, addressing challenges in AI research and integration. Llama Nemotron models offer high accuracy for various tasks. Project R2X, a vision-enabled PC avatar, demonstrates the capabilities of NIM microservices. APIs for different model types enhance AI innovation on PCs.

Developers can create AI-powered projects quickly using NVIDIA AI Blueprints, which offer reference implementations for complex workflows. Blueprints include everything needed for customization and extension. Two AI Blueprints for RTX include PDF to podcast and 3D-guided generative AI, offering artists greater control over image generation.

GeForce RTX 50 Series GPUs are designed for generative AI, featuring fifth-generation Tensor Cores with FP4 support. These GPUs offer faster G7 memory and an AI-management processor for efficient multitasking. FP4 support enables better performance and smaller model sizes for PCs, enhancing generative AI capabilities. NVIDIA introduces FP4, a memory-saving method that offers over 2x performance compared to FP16 on GeForce RTX 50 Series GPUs. Advanced quantization methods by NVIDIA TensorRT Model Optimizer ensure virtually no loss in quality.

Black Forest Labs’ FLUX.1 [dev] model at FP16 requires over 23GB of VRAM, limiting support to GeForce RTX 4090. With FP4, it only needs less than 10GB, making it compatible with more GeForce RTX GPUs.

GeForce RTX 4090 with FP16 generates images in 15 seconds, while GeForce RTX 5090 with FP4 can do it in just over five seconds. AI APIs for PCs, NVIDIA NIM microservices, and AI Blueprints will be available next month.

NIM-ready RTX AI PCs will be offered by Acer, ASUS, Dell, and more. GeForce RTX 50 Series GPUs promise game-changing performance and transformative AI experiences for creators. Watch NVIDIA CEO Jensen Huang’s keynote at CES for more AI news.



Read more at NVIDIA: Unveiling NIM Microservices and AI Blueprints