Next-gen Nvidia GPU "Rubin" is ahead of schedule, uses 3nm manufacturing and HBM4

Nvidia's Rubin GPUs are ahead of schedule, using 3nm process and HBM4, set for 2025 release.

: Nvidia's Rubin GPU series, employing TSMC's 3nm N3 process node and HBM4 memory, is anticipated to launch in the second half of 2025. Originally planned for 2026, the release was expedited due to demand and production enhancements, contrasting the earlier delays of the Blackwell series. While GB300 might explore a socketed design, performance expectations remain high with FP4 introduction.

Nvidia's upcoming Rubin GPU series, named after Vera Rubin, is expected to release earlier than planned, now aiming for the second half of 2025. These chips will utilize TSMC's advanced 3nm N3 process and incorporate HBM4 memory, promising significant performance boosts compared to their predecessor, Blackwell.

The Blackwell GPUs, encountering design flaws and overheating issues, faced delays but are nevertheless sold out through 2025, purchased heavily by tech giants like Microsoft, Google, and Meta. Supply challenges with TSMC's CoWoS-L technology contributed to these delays, but production is expected to expand with Rubin's launch.

As Nvidia accelerates Rubin's schedule, competitors like AMD and Intel are developing MI300 and Falcon Shores, respectively. The GB300, formerly Blackwell Ultra, might adopt a socketed design to streamline production, though Nvidia is keen on avoiding performance sacrifices.

Nvidia3nm processAI hardware