NVIDIA RTX Blackwell With A Beastly 96GB Of GDDR7 RAM Breaks Cover

While we cannot confirm the name of this graphics card posted on X, it is likely that it will be part of an upcoming Blackwell-based RTX 6000 or RTX 8000 Blackwell series. According to the manifest, the card uses a PCB numbered PG153, which has not been used for any of the other cards that NVIDIA has already announced. This graphics card will most likely target professional workstations for content creation, data sciences, and visualization.

Other than the memory configuration, bus width and PCB numbering, there's not much explicitly shown on the manifest. But since the bus with is 512-bits wide, we can infer that the card is most likely powered by the GB202 GPU, which is the same chip powering the GeForce RTX 5090. It is possible, however, than any professional workstation card based on the GB202 will have additional cores enabled, to go along with the much larger frame buffer. A we noted in our GeForce RTX 5090 review, the GB202 employed on the card isn't fully enabled.
A full GB202 GPU includes 12 Graphics Processing Clusters (GPCs), 96 Texture Processing Clusters (TPCs), and 192 Streaming Multiprocessors (SMs) each with 128 CUDA Cores. As such, a full GB202 GPU includes 24,576 CUDA Cores, 192 RT Cores, 768 Tensor Cores, and 768 Texture Units. The GeForce RTX 5090, however, has “only” 11 GPCs, 85 TPCs, and 170 SMs enabled. Since NVIDIA's pro-vis workstaion cards often feature more cores than their gaming-oriented counterparts, it's possible this card will feature a fully-enabled GB202.
We should know for sure once the GeForce RTX 50 family is fully launched and NVIDIA shift their focus to its next-generation professional workstation cards.
Top image source: NVIDIA