RTX 5090

The RTX 5090 is a graphics card made by the company Nvidia. A graphics card is a component in a computer that carries out huge numbers of simple calculations at the same time. That is exactly what is needed to compute moving images in games. The RTX 5090 came onto the market in early 2025 and was, at that point, the fastest model that private individuals could buy. The price was around 2000 euros, though in the first months it was often noticeably higher due to shortages. It belongs to the so-called Blackwell generation, a particular series of Nvidia chips.

Why a gaming card shows up in AI news

Images in games and AI models require the same kind of calculation. In both cases, huge tables of numbers are multiplied with one another. A normal main processor does this one step at a time and is slow at it. A graphics card has thousands of small computing units that work in parallel. That is why it is hundreds of times faster for AI.

What matters most with the RTX 5090 is its own on-board memory. It has 32 gigabytes of it, considerably more than most previous models. An AI language model has to fit completely into this memory, otherwise it either won’t run at all or will run extremely slowly. With 32 gigabytes, you can get mid-sized models running locally. That is precisely why developers and researchers buy this card, even though it is sold as gaming hardware.

A second reason is political. The USA restricts the export of particularly powerful AI chips to China. Cards like the RTX 5090 therefore regularly appear in news stories about smuggling and export rules. A graphics card for video games has thus become an object of trade policy.

What’s inside the card

The centerpiece is a single large chip with over 20,000 computing units. Nvidia calls them CUDA cores. Each one is simple, but they all work simultaneously. Alongside them sit specialized components called Tensor cores. These are built solely for the number tables used by AI models and speed them up even further.

The card’s memory is of the GDDR7 type and connected at very high speed. This matters more than it sounds. With AI applications, the bottleneck usually isn’t computing power but rather reloading numbers from memory. Think of it like a kitchen: the cooks are fast enough, but the ingredients arrive too slowly.

The price for all of this is power consumption. Under full load, the RTX 5090 draws up to 575 watts, more than a small space heater. It needs a strong power supply, a large case, and good ventilation. It simply doesn’t fit into many pre-built office computers.

Between gaming PC and data center

In everyday life, most people encounter the card in game benchmarks. These focus on frame rates in current titles and on a technology called DLSS, in which AI generates intermediate frames instead of fully computing them. Video editing and 3D software also benefit significantly. For pure gaming, however, it is overkill — cheaper models are almost always sufficient.

The second group consists of people who want to run AI models on their own computer. The advantage is independence: no data leaves the computer, and there are no ongoing fees. A common misconception, though, is that this lets you recreate ChatGPT. The largest models require the combined memory of dozens of such cards.

Data centers therefore use different products, such as Nvidia’s B200. These cost a multiple of the price per unit and are designed to work together by the thousands. The RTX 5090 is the most powerful card for a single desk, not for a server hall. Still, it turns up in stock market news because it shows how closely the gaming market and the AI business have become intertwined.