NVIDIA’s next-generation Rubin GPU architecture will enter “trial production” in 2H 2025 according to the latest rumors, with SK hynix working on getting its next-gen HBM4 memory ready earlier than expected.
VIEW GALLERY – 2 IMAGES
The company promised a 1-year cadence on new AI GPU architecture releases, but it seems NVIDIA is even more keen on getting new AI GPU architectures into the wild even quicker. In a new report from ZDNet Korea, we’re learning that SK hynix is shipping out HBM4 AI memory samples in June at the latest, with mass production of its new HBM4 expected to begin in Q3 2025, around 3 months earlier than SK hynix had planned.
NVIDIA requested SK hynix bring its HBM4 AI memory to market ahead of schedule, and this is ahead of the expedited release of HBM4 by NVIDIA. ZDNet Korea reports that the tape-out of HBM4 to NVIDIA had taken place in Q4 2024, which means SK hynix has already completed verification stages with big partners.
NVIDIA will also reportedly call SK hynix its “exclusive” HBM4 customer, meaning that NVIDIA will get its hands on the bleeding-edge new memory far earlier than its competitors. We should expect a next-generation Rubin R100 AI GPU release with ultra-fast HBM4 memory in the second half of 2025 because of this.
Blackwell B200 and GB200 are making their way out into the market, with its new “market-grabbing weapon” in new GB300 AI GPUs to get unveiled at NVIDIA GTC 2025 in March, and now we’ve got teases of the new Rubin R100 AI GPU coming with HBM4 earlier than expected. Where you at, AMD? Intel?