It appears that the NVIDIA RTX “Blackwell” GPU has 96 GB of GDDR7 memory on a 512-bit bus.
According to recent shipping manifests, NVIDIA requires a graphics card with 96 GB of GDDR7 memory to function. According to documents, the product uses a clamshell design, which combines two 3 GB modules per memory controller, and a 512-bit memory bus. The memory capacity of current workstation-oriented GPUs is essentially doubled with this configuration. The GB202 chip, the only Blackwell desktop GPU with a 512-bit interface, is thought to be used in the product. A board with the identification PG153—which is absent from all of NVIDIA’s current consumer GPUs—is mentioned in the docs. This result suggests a workstation or professional model instead of a gaming product. Additionally, it can be a member of the RTX 8000 Blackwell or 6000 Blackwell series.
The RTX 6000 “Ada,” the highest workstation card from NVIDIA right now, has 48 GB of RAM. A significant increase to 96 GB would allow for more intricate workloads for AI, data analysis, and content production. The power target of this GPU may be substantially higher than that of the workstation variants available now. However, in order to keep power consumption within reasonable bounds and enable more reliable operation in professional settings, professional GPUs frequently retain lower clock speeds. The exact name of the card and its final characteristics, like its core count and actual clock frequencies, have not been confirmed. In the past, NVIDIA’s professional GPUs have offered more cores than their gaming equivalents. The speculated 96 GB GPU might even outperform the GeForce RTX 5090, which has 32 GB of GDDR7, if it continues on this trajectory. In March, NVIDIA is anticipated to host its yearly GPU Technology Conference. Official announcements are thought to be likely to take place at this occasion. These details are still unconfirmed until then.
