After a breaking records 2024Nvidia is kicking off 2025 with a bang, unveiling a slew of products that could cement its dominance in AI and gaming development.
CEO Jensen Huang took the stage at CES in Las Vegas to show off new hardware and software offerings that include everything from personal AI supercomputers to next-generation gaming cards.
Nvidia’s biggest announcement yet: Project DIGITS, a $3,000 personal AI supercomputer that packs a petaflop of computing power into a desktop-sized box.
Built on the new – and still secret – GB10 Grace Blackwell Superchip, this machine can process AI models with up to 200 billion parameters while drawing power from a standard wall outlet.
For larger workloads, users can connect two units and solve models with up to 405 billion parameters.
For context, the largest Llama 3.2 model, Meta’s most advanced open-source LLM, has 405 billion parameters and cannot be run on consumer hardware.
So far it has required about 8 Nvidia A100/H100 superchips, each costing around 30 thousand dollarstotaling over $240,000 just for hardware processing.
Two new consumer AI supercomputers from Nvidia would cost $6,000 and be able to run the same quantized model.
“AI will be mainstream in every application for every industry. With Project DIGITS, the Grace Blackwell Superchip is coming to millions of developers,” said Jensen Huang, CEO of Nvidia, in an official statement. blog post. “Putting an AI supercomputer on the desk of every data scientist, AI researcher and student empowers them to engage and shape the age of AI.”
For those who love technical details, the GB10 chip represents a significant technical achievement born from the collaboration with MediaTek.
The system-on-chip combines Nvidia’s latest GPU architecture with 20 energy-efficient ARM cores connected via NVLink-C2C interconnect.
Each DIGITS unit has 128GB of unified memory and up to 4TB of NVMe storage. Again, for context, the most powerful GPUs to date contain around 24GB of VRAM (the memory needed to run AI models), and the H100 Superchip starts at 80GB of VRAM.
Nvidia’s plans to dominate AI agents
Companies they rush to deploy AI agentsand Nvidia knows it, which is probably why developed by Nemotrona new model family that comes in three sizes, and today announced its expansion with two new models: Nvidia NIIM for video summarization and understanding, and Nvidia Cosmos, which provides Nemotoron vision capabilities – the ability to understand visual instructions.
Until now, LLMs have been text-only. However, the models excelled in the following instructions: chat, function calls, coding, and math tasks.
They are available through Hugging Face as well as Nvidia’s website with enterprise access via the company’s AI Enterprise software platform.
Again, for context, V LLM ArenaNvidia’s Llama Nemotron 70b is higher than the original Llama 405b developed by Meta. It also defeats various versions of Claude, Gemini Advanced, Grok-2 mini and GPT-4o.
Nvidia’s push agent is now also related to infrastructure. The company has announced partnerships with major agent technology providers such as LangChain, LlamaIndex and CrewAI to create plans for Nvidia AI Enterprise.
These ready-to-deploy templates solve specific tasks that make it easier for developers to create highly specialized agents.
A new PDF-to-podcast proposal aims to compete with Google’s NotebookLM program, while another plan helps create search and video summarization agents. Developers can test these plans through a new Nvidia Launchers a platform that enables one-click prototyping and deployment.
Players, rejoice! The new GeForce RTX 5000 cards are performance beasts
Nvidia saved its gaming announcements for last and revealed the much anticipated GeForce RTX 5000 series. The flagship RTX 5090 contains 92 billion transistors and delivers 3,352 trillion AI operations per second – double the performance of the current RTX 4090. The entire lineup includes fifth-generation Tensor Cores and fourth-generation RT Cores.
The new cards introduce DLSS 4, which can increase the frame rate up to 8x using AI to generate more frames per render. Blackwell, the AI engine, has arrived for PC gamers, developers and creatives,” Jensen Huang he said“Combining artificial intelligence-driven neural rendering and ray tracing, Blackwell is the most significant innovation in computer graphics since we introduced programmable shading 25 years ago.”
The new cards also use transformer models for super-resolution, promising highly realistic graphics and much more performance for their price – which is not cheap, including: $549 for the RTX 5070, with the 5070 Ti for $749, the 5080 for $999, and the 5090 for $1,999.
If you don’t have that kind of money and want to play, don’t worry.
AMD too she announced today its Radeon RX 9070 series. The cards are built on the new RDNA 4 architecture using the 4nm manufacturing process and are equipped with special AI accelerators that compete with Nvidia’s tensor cores.
While the full specs remain under wraps, AMD’s latest Ryzen AI chips already hit the TOP 50 for peak performance.
Sadly, this still makes Nvidia the king of AI applications CUDA technologyNvidia’s proprietary AI architecture.
To address this issue, AMD has secured partnerships with HP and Asus for system integration, and more than 100 enterprise platform brands will use AMD Pro technology by 2025.
Radeon cards are expected to hit the market in Q1 2025, giving Nvidia an interesting battle in both gaming and AI acceleration.
Edited Sebastian Sinclair
Generally intelligent Bulletin
A weekly AI journey narrated by Gene, a generative AI model.
