Nvidia R100 GPU with HBM4 memory will reportedly enter mass production in Q4 2025

By now, you might've already heard that Nvidia's new GPU architecture will be named after the renowned astrophysicist Vera Rubin. Although Nvidia has yet to confirm the existence of this new architecture, there have been new rumours regarding its specs.

Earlier this year, Nvidia unveiled its latest and most potent AI chip, the Blackwell B100, which is slated for release later this year. However, some people are already looking forward to the company's future products and what they'll offer. According to analyst Ming-Chi Kuo (via VideoCardz), the R100 GPU is expected to go into mass production by Q4 2025, with the system and rack solutions starting no earlier than Q1 2026.

Sources suggest that Nvidia is still finalising the specifications, especially the interposer size and packaging. The company is said to be considering three options, including a 4x reticle design that uses CoWoS-L packaging, also used in Blackwell GPUs. The Vera Rubin architecture is expected to employ the TSMC N3 process node and pack four HBM4 stacks.

Rumour has it that Nvidia is focusing on power consumption as one of the key areas of the Vera Rubin architecture. With the B200 GPUs consuming up to 1000W, we're curious about how much power Rubin GPUs will consume.

Discuss on our Facebook page, HERE.

KitGuru says: It is unclear if Rubin will be used for anything other than data centre GPUs, so we'll have to wait and see. There is also plenty to learn about the features and performance capabilities of this upcoming architecture. 

Become a Patron!