In these days various rumors emerged which outline what could be the technical characteristics of the future generations of video cards that NVIDIA will make GeForce available in its family of products. So let’s talk about the cards GeForce RTX 40, models that they shouldn’t see the light before the second half of 2022 following the 2-year cadence that separates the launch of new architectures by the American manufacturer.
The GeForce RTX 3000 solutions currently available on the market are based on the architecture known as Ampere. This will happen to that Lovelace, intended in the intentions of the American company to ensure a further leap forward in terms of performance with 3D applications. There are rumors of a top-of-the-range proposal equipped with 144 streaming multiprocessors internally, which correspond to 18,432 CUDA cores. By way of comparison, we recall that the GeForce RTX 3090 cards integrate a GPU with 10,496 CUDA cores collected in 82 streaming multiprocessors.
The Lovelace proposals will clash with the next generation of video cards developed by AMD: let’s talk about those GPU-based Radeon with Navi 3 architecture, for which the rumors currently available speak of a approccio MCM (Multi Chip Module) with multiple GPU modules side by side mounted on the same package. This would in fact be the same type of approach adopted for Ryzen processors, which in the versions with a higher number of cores are built alongside more die each with 8 cores.
NVIDIA also seems willing to adopt this approach but will do so, according to the information currently available, only with the proposals based on architecture Hopper which will arrive on the market after the Lovelace one. From this some speculations that open the possibility that NVIDIA can bring the Hopper architecture also in the consumer solutions segment, against an initial positioning expected only for the professional market, so as to be able to better compete with the AMD proposals if the cards with Navi 3 GPU turned out to be particularly brilliant in overall performance.
In our opinion, this is unlikely as the general plans, in addition to the chip design phase, for the solutions with Lovelace architecture have now certainly been defined internally by NVIDIA. It also speaks of 5 nanometer manufacturing technology for Lovelace GPUs, with production by the Taiwanese TSMC even if it is not known at the moment whether with the N5 process or with the N5P one.
A picture therefore overall a bit confused, as it must be when it comes to unofficial information referring to products that at best we will see at the debut on the market in no less than 1 year.