features, prices, DLSS 4, what you need to know

As expected, Nvidia unveiled its new generation of GeForce RTX 5000 graphics cards during its CES 2025 conference. Here’s everything you need to remember.

Blackwell, DLSS 4, Reflex 2: the new features

No more Ada, here’s Blackwell. The GeForce RTX 5000 uses a new architecture from Nvidia. On the program, 5th generation Tensor Cores, 4th generation RT Cores, for ever more AI.

Nvidia is notably introducing RTX Neural Shaders, which allows the use of AI through programmable shaders. This will require developers to evolve video games, but it could allow a big leap in terms of photorealism.

For content creators, Blackwell also introduces a new NVENC video encoding engine and a new video decoding engine.

DLSS 4

This computing power dedicated to AI will, for example, be used by DLSS 4. This is a major overhaul of DLSS which shifts to technology based on transformers (the same technology that powers generative AI). modern like ChatGPT).

DLSS 4 therefore promises to improve all aspects of DLSS: uspacling (Super Resolution), Anti-Aliasing (DLAA), ray tracing (Ray Reconstruction) and Frame Generation. These new features will be offered to all RTX 2000, 3000 and 4000 (Frame Generation remains exclusive to the 4000 and 5000).

It takes the opportunity to add a new layer, reserved for the RTX 5000, Multi Frame Generation. The possibility for a game to generate several intermediate images entirely by AI to streamline the action. Nvidia promises to increase the performance of a game by 4 or 8 thanks to this.

Nvidia Reflex 2

Nvidia Reflex has been integrated into many competitive games since 2020. It significantly reduces latency between player actions and the reaction of the image to the screen, by optimizing the process from the mouse to the screen passing by the CPU and the GPU.

With Reflex 2, Nvidia promises to pass a new gap in terms of latency reduction. The firm wants to anticipate the rendering of the image based on the player’s last mouse movement, just before the image is sent from the GPU to the screen. A last (mili)second change that reduces the latency between the player’s movement and the reaction to the image.

The result is a latency promised to be further divided by two. By combining Reflex 1 and Reflex 2, Nvidia promises that a game can go from 56 ms of latency to just 14 ms.

Nvidia Reflex 2 will be offered on all GeForce RTX 2000, 3000, 4000 and 5000 graphics cards, with temporary exclusivity for the latter.

Graphics cards

Nvidia introduces its new range with four models: RTX 5090, RTX 5080, RTX 5070 Ti and RTX 5070.

GeForce RTX 5090

• GPU GB202
• 21,760 CUDA cores
• 32 Go de GDDR7
• 512-bit memory bus
• Power required: 575W

GeForce RTX 5080

• GB203
• 10,752 CUDA cores
• 16 Go de GDDR7
• 256-bit memory bus
• Power required: 360W

GeForce RTX 5070 Ti

• GB203
• 8960 CUDA cores
• 16 Go de GDDR7
• 256-bit memory bus
• Power required: 300W

GeForce RTX 5070

• GB205
• 6144 CUDA cores
• 12 Go de GDDR7
• 192-bit memory bus
• Power required: 250W

Launch and price

The GeForce RTX 5090 and 5080 will be launched in January, starting at 2349 and 1179 euros respectively.

The GeForce RTX 5070 Ti and RTX 5070 will be launched in February, starting at 884 and 649 euros.

Please note, however, that these are the prices announced by Nvidia for its Founder Editions. The reality among other manufacturers, and on the French market in particular, could be quite different.