The future GeForce RTX 5090 based on the Blackwell architecture would be the graphics card of all superlatives. Nvidia would check all the boxes to push a card’s computing power to new heights.
What is Nvidia working on with its next generation of graphics cards? The Internet user Panzerlied to whom we already owe indiscretions concerning Nvidia has published some assertions on the reference forum Chiphell. What he says has been corroborated by Kopite7kimianother name used to Nvidia leaks.
With the GeForce RTX 5000, Nvidia should switch to the Blackwell architecture. For the first time, it would be based on a chiplet design rather than a monolithic design, which should make manufacturing easier for Nvidia. If we believe the figures of Panzerliedthis should above all allow significant gains in computing power.
More than 24,000 cores?
According to the person concerned, the Nvidia GeForce RTX 5090 would integrate 50% more CUDA cores, compared to the GeForce RTX 4090 which would bring the total to 24,576 hearts, that’s 8x more than the GeForce RTX 2080 at the time. The frequency would also increase by 15% and could exceed the 3 GHz mark. Between the advances enabled by the architecture, the increase in the number of cores and the frequency, the GeForce RTX 5090 could be much more powerful than previous generations.
Bandwidth would also jump 50% to 32 Gb/s, suggesting the use of very modern GDDR7. The L2 cache would increase and would increase to 128 MB. The quantity of VRAM would increase to 32 GB on a 512-bit bus.
It is very early to draw conclusions about these rumors. Indeed, the GeForce RTX 5000 are not expected before 2025. Nvidia would give itself time to popularize its RTX 4000 generation and the Ada Lovelace architecture a little more. The firm could still take advantage of 2024 to improve its catalog with “Super” versions as was the case with the RTX 2000 generation.