NVIDIA supports generative AI with Tensor Core GPUs, LLM, and RTX PC and workstation tools.

NVIDIA announced the new GeForce RTX 40 SUPER GPU series with generative AI performance at CES 2024 in Las Vegas on the 9th. In addition, new AI laptops from major manufacturers, NVIDIA RTX™ accelerated AI software and tools were announced.

At CES 2024, NVIDIA also announced the launch of new generative AI-based applications and services for PCs. This includes: NVIDIA TensorRT acceleration of the popular Stable Diffusion XL model, which supports text-to-image workflows; NVIDIA RTX Remix with generative AI texture tools; games using DLSS 3 Frame Generation; and NVIDIA ACE microservices.

“Generative AI is the most significant platform shift in the history of computing, and it will transform every industry, including gaming,” said NVIDIA CEO Jensen Huang. “NVIDIA’s RTX GPUs, with more than 100 million units shipped, are already playing a pivotal role in generative AI, such as ChatGPT and Stable Diffusion,” said Jeff Fisher, senior vice president of GeForce at NVIDIA.

Meanwhile, last October, NVIDIA released the TensorRT-LLM library for Windows, accelerating large language models (LLMs) such as Rama 2 by up to 5x on RTX PCs. Later this month, it plans to support RTX-accelerated LLMs from devices to YouTube videos through the Chat with RTX playground.

With 'Chat RTX', users are expected to be able to personalize LLMs linked to their content, such as documents, notes, videos, and other data. It is explained that they can quickly get contextual answers with a customized chatbot by utilizing Augmented Search Generation (RAG), TensorRT-LLM, and RTX acceleration. Additionally, open source reference projects are provided so you can easily implement the same functionality in your own applications.

■ GeForce RTX 40 Super Series 3 Types Unveiled


The 'GeForce RTX 40 SUPER' series announced at CES 2024 includes three models: 'GeForce RTX 4080 SUPER', '4070 Ti SUPER', and '4070 SUPER'.

Of the three products released this time, the 'GeForce RTX 4080 Super' is 1.4x faster than the non-frame-generating RTX 3080 Ti in graphically intensive games such as 4K. The RTX 4080 Super has more cores and faster memory. It will be released on January 31st and is priced at $999.

The 'RTX 4070 Ti Super' adds more cores, increases the frame buffer to 16GB, and increases the memory bus to 256-bit. It is 1.6x faster than the 3070 Ti, and 2.5x faster at DLSS 3. The RTX 4070 Ti Super will be available on January 24th for $799.

The RTX 4070 Super adds 20% more cores, making it faster and more power-efficient than the RTX 3090. With DLSS 3, it’s up to 1.5x faster in the most demanding games. The product will be available starting January 17th for $599.

Manufacturers including Acer, ASUS, Dell, HP, Lenovo, MSI, Razer, and Samsung are releasing RTX AI laptops, putting the full set of generative AI capabilities into users’ hands immediately. The new systems, which deliver “ranges from 20x to 60x performance gains over NPUs [neural processing units],” NVIDIA said, are launching this month.

The mobile workstations, which feature RTX GPUs, include a three-year license with the new release to enable streamlined and secure generative AI development with AI software such as TensorRT and RAPIDS™.

■ Providing PC developer tools for building AI models

NVIDIA is offering a new PC developer tool for building AI models. The AI Workbench is a unified toolkit for AI developers that allows them to quickly create, test, and customize LLMs using PC-class performance and memory space. It will be available in beta later this month.

The AI Workbench provides efficient access to repositories like Hugging Face, GitHub, and NGC™. It’s simplified for developers to easily reproduce, collaborate, and migrate projects. They can scale projects to data centers, public clouds, NVIDIA DGX™ Cloud, and more, then bring them back to their local RTX systems on their PCs or workstations for inference jobs.

NVIDIA partners with HP to simplify AI model development. Integrates NVIDIA AI-based models and endpoints into HP AI Studio, allowing users to easily search, import, and deploy optimized models across PCs and clouds. Developers can also build AI models for PC use cases and then optimize them with NVIDIA TensorRT to take full advantage of the Tensor Cores in RTX GPUs.