Close up photo of RTX 5090 graphics card against black background. Dramatic studio lighting highlighting the cooling fans and metal shroud. Shot on RED camera with 85mm lens, f2.8, shallow depth of field.
Created using Ideogram 2.0 Turbo with the prompt, "Close up photo of RTX 5090 graphics card against black background. Dramatic studio lighting highlighting the cooling fans and metal shroud. Shot on RED camera with 85mm lens, f2.8, shallow depth of field."

NVIDIA Doubles GPU Speed with RTX 50 Series at CES 2025

Jensen Huang just announced the most powerful consumer GPU ever made. The RTX 5090 packs twice the speed of the 4090, with 92 billion transistors pumping out over 3,352 trillion AI operations per second.

But raw power isn’t the full story. NVIDIA built the RTX 50 Series around their new Blackwell architecture, which brings major upgrades to both gaming and AI tasks. The RTX 5070 matches the 4090’s performance at $549, while the 5070 Ti hits $749.

DLSS 4.0 takes AI-powered rendering to another level, predicting entire frames and pixels to boost visuals without killing performance. For data centers, the Blackwell GB200 GPUs deliver 4x better efficiency than the last generation, connected through NVLink 72 into massive AI processing farms.

NVIDIA isn’t stopping at traditional computing either. Their new Cosmos World Foundation Model, trained on 20 million hours of video, lets AI understand and predict real-world physics. Combined with Omniverse, this powers everything from robotic warehouses to self-driving cars.

Speaking of autonomous vehicles, NVIDIA announced a partnership with Toyota using their new Thor chip – 20 times faster than the current Orin processor. They’re building a complete stack with DGX for training, Omniverse for simulation, and AGX for in-car AI.

For developers and AI researchers, Project DIGITS introduces a personal AI supercomputer built around the GB110 chip. This compact wireless system brings data center capabilities to your desk when it launches in May 2025.

I’ve written before about the rapid progress of AI hardware in my post on AI predictions for 2025 (https://adam.holter.com/ai-predictions-2025-agents-video-and-open-source-models/). This announcement from NVIDIA proves those predictions might have been conservative.

The real impact here isn’t just better gaming or faster AI training. NVIDIA has built an end-to-end system spanning consumer PCs, enterprise AI, robotics, and autonomous vehicles. They’re not just making faster chips – they’re building the foundation for AI to move from data centers into the physical world.