IT Home reported on June 10 that the Taiwanese media "Business Times" reported yesterday that the development process of Nvidia's next-generation AI GPU "Vera" and CPU "Rubin" was smoother than before. Tape-out was completed this month, and samples were released to customers as early as September, and mass production was carried out in early 2026.
Nvidia has previously stated that it is based on the "Rubin" GPU and the "Vera" CPU The Vera Rubin NVL144 rack-level system will be launched in the second half of next year, with performance up to 3.3 times that of the previous generation GB300 NVL72.
A "Rubin" GPU will contain 2 mask-sized GPU chips, and it is reported that TSMC's N3P process will be used; and the "Rubin" GPU and 8 HBM4 memory will also be integrated with TSMC's CoWoS-L advanced packaging technology. In addition, the I/O chip used in this system will also be based on TSMC process.
