Nvidia launches the GeForce RTX 50 series and immediately offers a full range of GPUs. Based on the Blackwell architecture, the new graphics cards should appeal to professionals and gamers.
A little over two years after the introduction of the RTX 40, Nvidia is ready to move up a gear. At CES, the GPU specialist is presenting the GeForce RTX 50, a family of GPUs built around the same Blackwell architecture that today powers the most powerful AI accelerators in the world. At the high end, Nvidia reigned supreme for two years with the Ada Lovelace-based RTX 40 series, and with Blackwell RTX 50, CEO Jensen Huang wants to repeat the feat.
Integrated AI
As expected, the Nvidia RTX Blackwell has the necessary AI capabilities. Nvidia merges the computing power of GPU and AI into a tandem that works together to display high-resolution images at high frame rates. The new GPUs support the fourth generation of DLSS, in which a neural network can generate up to eight times more frames than with traditional rendering. This allows you to increase the rendering time per frame, which allows you to obtain sharper scenes thanks to raytracing, for example.
With the RTX 50, Nvidia is targeting not only lighting, but also textures. AI generation allows relatively low resolution textures to be loaded, after which the real-time AI takes care of the upscaling.
Plus, this latest GPU line has a mountain of extra graphics power under the hood. The high-end Nvidia GeForce RTX 5090 is said to be twice as fast as the GeForce RTX 4090 in terms of image generation, thanks in part to its integrated AI capabilities.
Twice as fast
Nvidia is launching a total of four new GPUs, each of which is expected to be around twice as fast as their predecessors. The following desktop cards will appear on shelves in the coming weeks:
| Nvidia GeForce | CUDA Cores | Tensor (AI-Tops) | Ray tracing (TFlops) | GHz (Boost/base) | GDDR7 (GB) | Interface | Retail price (dollars) |
| RTX 5090 | 21.760 | 3.352 | 318 | 2,41 / 2,01 | 32 | 512 bits | 1.999 |
| RTX 5080 | 10.752 | 1.801 | 171 | 2,62 / 2,3 | 16 | 256 bits | 999 |
| RTX 5070 Ti | 8.960 | 1.406 | 133 | 2,45 / 2,3 | 16 | 256 bits | 749 |
| RTX 5070 | 6.144 | 988 | 94 | 2,51 / 2,16 | 12 | 192 bits | 549 |
As you can see, the RTX 5090 is an ultimate powerhouse with twice the horsepower of the RTX 5080 and an equivalent price. Other GPUs are still very powerful. The amount of onboard memory is great for gamers, but with the exception of the 5090, it’s not fully optimized for pure AI workloads. For example, to let in large models, memory is sometimes more important than GPU performance. This is what AMD is banking on, for example, with its Ryzen AI Max APUs.
Also for laptop
Nvidia is launching the RTX 50 not only for desktops, but also immediately for laptops. The laptop variants have similar names to their desktop counterparts, but are much less powerful. A laptop RTX 5090, for example, has specifications closer to a desktop RTX 5080. You can clearly see the difference below:
| Nvidia GeForce | CUDA Cores | Tensor (AI-Tops) | GDDR7 (GB) | Interface | TDP (W) |
| RTX 5090 Laptop | 10.496 | 1.824 | 32 | 256 bits | 95-15 |
| RTX 5080 Laptop | 7.680 | 1.334 | 16 | 256 bits | 80-150 |
| RTX 5070 Ti Laptop | 5.888 | 992 | 16 | 192 bits | 60-112 |
| RTX 5070 Laptop | 4.608 | 798 | 12 | 128 bits | 50-100 |
Laptop models have varying TDP, so the actual clock frequency and performance will vary depending on the laptop manufacturer’s implementation. Laptops powered by the RTX 50 will be available starting in March this year.
From data center to desktop
Nvidia is competing again with the RTX 50. Huang and his team have transformed in recent years. While not so long ago, Nvidia was a GPU specialist unleashing its knowledge on AI accelerators, today it’s the opposite. Blackwell was developed for AI training in data centers, and from there, Nvidia brought the architecture to consumers.
For workstations aimed at graphics professionals, the RTX 50 promises to give a big boost, even if it is mainly gamers who will be interested in these cards initially. We’re curious to see actual availability and pricing: Nvidia GPUs have a habit of being a bit more expensive and less in stock shortly after launch than the general public might want.
