Nvidia will release 16 versions of GeForce Blackwell RTX 5000-series graphics cards

On the eve of the announcement of the new generation of Nvidia GeForce RTX 5000 video cards, based on the Blackwell architecture, information about the new models appeared in the PCI Device ID database. Although this is only a list of potential models without exact specifications, it gives some idea of ​​the products in the pipeline. Currently, there are three SKUs for desktop graphics cards and more than ten SKUs for mobile solutions.

For laptops, a wide range of graphics cards are expected, starting from the top-of-the-line GeForce RTX 5090/5080 Max-Q/Mobile variants and ending with budget models such as the GeForce RTX 5050. All mobile versions will be presented in two variants with different SKUs. Nvidia now labels all mobile graphics cards as Max-Q, regardless of performance level, leaving it up to laptop manufacturers to set power limits.

According to rumors, the next generation GeForce RTX 50 will be announced at CES 2025 in January. Nvidia is expected to introduce mobile and desktop graphics cards at the same time or with a slight gap in time.

Previously Nvidia confirmed its participation in CES 2025, and the company’s CEO, Jensen Huang, will deliver a keynote speech on January 6. At this event, the announcement of new graphics processors of the GeForce RTX 50 series is expected, including the flagship models RTX 5090 and RTX 5080.

The entire line will have 14-layer printed circuit boards, they will provide improved signal integrity between the GPU and GDDR7 video memory. All cards must have a single 12V-2×6 power connector interface. Only some top-end RTX 5090 or RTX 5080 models will have a dual 16-pin connector.

Nvidia GeForce RTX 5090 is based on GB202-300-A1 GPU and PG144/145-SKU30 PCB. The GPU contains 170 active stream multiprocessors (out of a total of 192 SMs) and 21,760 cores (out of a total of 24,576 cores). It has 32 GB of GDDR7 memory with a 512-bit access bus. The memory works at a speed of 28 Gbit/s and provides a bandwidth of 1792 GB/s. The total power consumption of the card is 600 W.

Nvidia GeForce RTX 5080, in turn, will be equipped with 16 GB of VRAM, targeting a wider range of users. The access bus was also halved to 256 bits. However, it uses faster GDDR7 memory (32 Gbit/s), which provides a total bandwidth of 1024 GB/s. The model is based on the PG144/147-SKU45 board and uses the GB203-400-A1 GPU. In this case, all 84 thread multiprocessors and 10752 cores available in the GPU are used, but this is 51% less than in the flagship RTX 5090. In comparison, the previous RTX 4080 had 40% fewer function blocks than the RTX 4090.

Nvidia GeForce RTX 5070, according to the latest leaks, will be based on the PG147 board and GB205 GPU. Exact specifications regarding the number of function blocks are still unknown, but it is assumed that it will receive 12 GB of memory with a 192-bit bus, just like its predecessor RTX 4070. However, thanks to the use of faster memory with a bandwidth of 28 Gbps, the total the bandwidth of the video card will increase by 33% to 672 GB/s, which will significantly increase its performance. In terms of power consumption, the GeForce RTX 5070 will require 250W, which is 14% more than the RTX 4070 SUPER (220W).

There are also rumors that Nvidia may introduce other models in the RTX 50 series, including entry-level variants, but the launch of the laptop GPUs is likely to be delayed. This decision may be related to weakened competition from AMD, which gives Nvidia time to improve its new products.

AMD, for its part, plans to present the new Radeon RX 8000 series video cards at CES 2025. Despite the interest in these products, they do not yet pose a significant threat to Nvidia’s market dominance.