R100 will form the basis of accelerators for AI
Nvidia only recently introduced the Blackwell architecture and accelerators based on it, they haven’t even hit the market yet, and information has already appeared on the Internet that the new generation will be shown next year. < /p>
TF International Securities analyst Mich-Chi Kuo says that mass production of the Rubin generation R100 GPU will begin in the fourth quarter of 2025, and production of the DGX and HGX servers will start in the first half of 2026. That is, customers will receive such products approximately by mid-2026.
R100 will be produced using a 3 nm process technology, which will significantly improve energy efficiency. The new GPU will receive eight stacks of the new HBM4 memory, the production of which Samsung promised to begin at the end of next year.
Nvidia will also release the Grace Superchip GR200 adapter with two R100 GPUs and a new Grace processor, which will also be produced according to 3 nm standards.
Recall that the Rubin generation will be named after the astronomer Vera Rubin, who made many discoveries. In particular, it was her observations of the discrepancy between the actual rotation rates of galaxies and the predicted ones that led to the creation of the theory of dark matter.