The GB202 “Blackwell” gaming GPU uses the same node as the B100 for data centers

Yesterday, NVIDIA unveiled its latest Blackwell GPU architecture, designed specifically for data centers and AI acceleration, marking the first step in its new architectural strategy. The company is reportedly planning to launch a new series of gaming GPUs built on the Blackwell foundation later this year. Interestingly, the new B100 GPU will not deviate much from its predecessor when it comes to processing nodes, instead relying on 4NP. This corresponds to an enhanced version of the 4N, a custom TSMC node originally developed for the Hopper series.

According to Kopite7kimi, NVIDIA will not make any changes to its gaming GPU series, which is also based on this node. This new node is still said to provide a 30% increase in transistor density. This means that both Ada and Blackwell will use a different version of the 5nm node. The leaker added that the Blackwell GB202 gaming GPU will see a change in the L1 cache (not to be confused with the L2 cache), which according to the leaker means a significant improvement over the AD102 and GA102, both of which have a large size. Than you have 128 KB. Such a change may increase the performance of a single stream multiprocessor.

Image source: videocardz

The GB202 GPU is expected to be the flagship of the RTX 50 series. Speculations suggest that it may feature 192 stream multiprocessors (SMs) and a powerful 512-bit memory bus. Given the dual architecture of the B100, it is plausible that a similar approach could also be applied to the smaller GB202 gaming processor.

More rumors suggest that the GB203's specifications could be around half of what the GB202 offers, meaning it could have around 96 SMs and half the transverse memory buses compared to the GB202 model.

The specifications of Blackwell's remaining gaming GPUs are still unclear, but it has been indicated that all RTX 50 series will support PCIe Gen5 and DisplayPort 2.1 and will feature GDDR7 memory.