NVIDIA unveils RTX 40 SUPER cards and brings generative AI to PCs

NVIDIA has opened CES 2024 with three new GeForce RTX 40 SUPER graphics cards and a clear ambition: to make generative artificial intelligence less dependent on remote servers. The new GPUs boost gaming performance, but they are also designed to accelerate AI models on personal computers.

Jensen Huang, NVIDIA’s founder and CEO, introduced the RTX 4080 SUPER, RTX 4070 Ti SUPER and RTX 4070 SUPER. The launch comes as models capable of generating text, images, voice and interactive characters begin to demand more computing power outside data centers as well.

Three graphics cards to refresh the high-end lineup

The GeForce RTX 4080 SUPER will launch on January 31 for $999, $200 less than the original RTX 4080’s launch price. NVIDIA has increased its CUDA core count—the units that perform parallel calculations—to 10,240 and equipped the card with 16 GB of GDDR6X video memory. The company says it delivers twice the performance of an RTX 3080 Ti in games with ray tracing and DLSS 3 enabled.

The RTX 4070 Ti SUPER sits below it, launching on January 24 for $799. Its most significant change is not just speed: video memory increases from 12 GB to 16 GB, and the card adopts a 256-bit memory bus. That extra capacity matters for high-resolution gaming, but also for working with AI models and image-generation tools that need to load large amounts of data into the GPU’s memory.

The first to arrive will be the RTX 4070 SUPER, on January 17, priced at $599. It has 7,168 CUDA cores—20% more than the original RTX 4070—and retains 12 GB of memory. NVIDIA is positioning it as an option for 1440p gaming with AI features such as DLSS 3, its system for generating additional frames to improve smoothness.

The GPU is no longer just for drawing pixels

The most important part of the announcement for AI involves the so-called Tensor cores. These components specialize in the mathematical operations used by neural networks, allowing them to run artificial intelligence tasks much faster than a conventional processor.

NVIDIA wants to use them to make certain generative applications run directly on the computer. The benefits are practical: lower latency, more control over personal files and the ability to use some features without sending every query to the cloud. The limitation is just as clear: a home PC cannot compete with the massive GPU clusters that train giant models, but it can run versions optimized for specific tasks.

At CES, the company showed Chat with RTX, a demo application that lets users create a local assistant from documents, notes and other files. The program uses language models and can search for information within a private collection of data without that data leaving the computer. NVIDIA said it will be available as a demo in February for systems with GeForce RTX GPUs.

This approach points to an emerging category: personal assistants that work with the data stored on a device itself. They do not replace services such as ChatGPT or Gemini, which have much larger models and access to massive infrastructure, but they can be useful when privacy, immediate responses or offline work are priorities.

Video game characters that talk back

NVIDIA also brought its ACE microservices to CES. They are designed to create avatars and non-player characters, or NPCs, powered by generative AI. A microservice is an independent piece of software that handles a specific function; in ACE, these components can recognize speech, generate responses, synthesize speech or animate a character’s face.

The proposal combines technologies such as Riva for speech, NeMo for language models and Audio2Face, which synchronizes facial animation with audio. The demo aims to move beyond the prewritten dialogue found in many video games: instead of choosing from several predetermined responses, players could speak with a character and receive an answer generated in real time.

That does not mean every studio can simply add fully improvised characters. Developers still need to control the tone, narrative boundaries, computing costs and response quality. AI can make a conversation more flexible, but it cannot design a good game on its own.

A race to turn the PC into an AI platform

The SUPER refresh does not change the RTX 40 series’ Ada Lovelace architecture, but it reinforces NVIDIA’s commercial message: the graphics card has become a central component of consumer AI. The company already dominates the data-center accelerator market; now it wants that same technological advantage to translate into everyday applications for personal computers.

For people with a recent RTX card, the main interest will not be limited to replacing hardware. During 2024, it will be crucial to see how many applications make genuine use of local execution: image editors, video tools, assistants for personal files and video games are the most immediate candidates. The RTX 40 SUPER cards bring more power to the table; what remains to be seen is whether software can turn that power into features worth having on a user’s desktop.