Everything we know about Nvidia's GeForce RTX 50 series
Which GPU for RTX 50 series?
The GeForce RTX 50 series graphics cards will adopt an iteration of the B200 graphics chip aimed at professional markets and announced with great fanfare last March. You should expect to see several less powerful versions throughout the range – starting from a B202 for the most powerful model for the GeForce RTX 5090 up to the GB205 for the entry-level models (5060). Like the 40 range, the chips will be engraved with TSMC's 4N process – where some analysts predicted a jump to the 3N process at the start of the year.
Nvidia is slowly but surely ceasing production of the GeForce RTX 40 series
What would that represent in terms of concrete power? Lacking a benchmark, it is impossible to make a tenable equivalence. From the first echoes within the industry, each Blackwell GPU would be superior to its Ada Lovelace generation equivalent for everything related to the generation of ray-tracing effects. The thermal envelopes announced should be taken with a grain of salt… And we prefer to believe the statements of a Segotep representative, declaring in last October that even the RTX 5090 will only need one 12 + 4 pin connector (12VHPWR), not two. As a reminder, this connection and its implementation by manufacturers have caused a lot of ink to flow since the launch of the RTX 40, as Gamer Nexus recounts in a documentary very complete.
How much memory?
Finally, after Samsung's formalization of 3 GB GDDR7 memory chips, it seems logical to find them in the GeForce RTX 50 series. The Korean giant promises a speed of up to 42.5 Gb/s – theoretical speed nonetheless , the application on GPUs should cap around 32 Gb/s for the most efficient models, still according to the latest rumors. This would nevertheless represent a substantial gain compared to the current generation GDDR6X, limited to 24 Gb/s. These larger memory modules would logically open the door to greater VRAM (video memory) capacity for GPUs: the high-end GeForce RTX 5090 model would see a 26 GB iteration of VRAM arrive after the release of its first 16 GB version .
| VRAM | TGP | Price estimate | Exit horizon | |
|---|---|---|---|---|
| GeForce RTX 5090 | 32 GB GDDR7 | 600W | $1900 – $2100 | January 2025 |
| GeForce RTX 5080 | 16 GB GDDR7 | 450W | 1300$ | January 2025 |
| GeForce RTX 5070 Ti | 16 GB GDDR7 | 300W | 800$ | Current 2025 |
| GeForce RTX 5070 | 12 GB GDDR7 | 250W | 600$ | March-April 2025 |
| GeForce RTX 5060 | 6 GB GDDR7 | 200W | 500$ | Current 2025 |
No suspense, all the lights are green for a widespread release in the first part of 2025. It will probably be done in several stages: the big officialization during CES 2025 (January 7 to 10) of the GeForce RTX 5090 , 5080 and 5070with a release during the first quarter. For the 5070 Ti and 5060 models, you will probably have to wait a little longer – until April, following Nvidia's previous releases. As for laptop models, we have no information on this subject other than that the RTX Geforce 4050 mobile would continue well into 2025. To find out more about the portable market, you will have to wait for CES or even Computex.
What about DLSS4?
It is much easier to glean information about the hardware of the future RTX 50 series than about its software features; production lines do not have the same sense of secrecy as Nvidia. Nevertheless, we can already expect a more or less substantial improvement in Super Resolutionits AI algorithm for image reconstruction from a smaller sample, or the Frame Generation which allows you to double the frame rate in the game. Moreover, the big man Jensen Huang reaffirmed his vision on the implementation of AI in gaming, seeing the technology allowing the generation of objects and characters in the game In short, a lot of ambition, but nothing very concrete at the moment.
RTX Remix integrates ComfyUI which generates textures by AI. © Nvidia
However, Nvidia is currently working on RTX Remix in order to modernize old games like Portal or Half-Life 2. In this project, the ComfyUI tool allows textures to be generated using very precise prompts. This tool could be integrated directly into the DLSS 4 which will accompany the GeForce RTX 50 series.
Better, faster, strongerthe future GeForce RTX 50 graphics cards have been the source of various and varied speculations for several months already. We expect a lot from their performance in terms of ray-tracing – more than the leap in rasterization that Nvidia is putting aside with the Ada Lovelace generation of the GeForce RTX series 40. But with the Blackwell architecture, the GeForce RTX series 50 should put doubling down on features using artificial intelligence… And putting a part back in the machine for even more impressive DLSS 4 potential. See you at CES.
