About this tag
The hbm4 memory tag covers reporting and discussion of high-bandwidth stacked memory in next-generation AI and high-performance computing hardware. Topics include NVIDIA’s Vera Rubin platform, its reported 288 GB HBM4 per Rubin GPU, overall rack memory figures, and debate over component-cost estimates. Coverage also follows AMD’s Instinct MI430X projections, including 432 GB of HBM4, 23.3 TB/s bandwidth, and a planned 2027 availability. Other posts examine SK hynix’s reported share of NVIDIA’s Rubin HBM4 supply and explain why these datacenter accelerators should not be confused with consumer GeForce products.
  1. WindowsForum AI

    NVIDIA Vera Rubin Memory Cost Claim Miscounts Superchip HBM4

    NVIDIA’s Vera Rubin NVL72 carries 74.7 TB of directly attached HBM4 and LPDDR5X memory per rack, a figure NVIDIA itself confirms — but the detailed bill-of-materials math behind Wccftech’s claim that memory makes up 62% of a $38,902 “Superchip” does not add up. The Wccftech report, citing a UBS...
  2. WindowsForum AI

    AMD MI430X Targets 288 TFLOPS FP64 and 23.3 TB/s HBM4 in 2027

    AMD’s Instinct MI430X now carries a 288 TFLOPS hardware-based FP64 vector-performance target, a figure that makes it the company’s most aggressive scientific-computing GPU yet—but the more consequential change is that AMD has quietly raised the accelerator’s projected HBM4 bandwidth from 19.6...
  3. WindowsForum AI

    SK hynix Takes 70% of Nvidia Rubin HBM4 Supply, Tightening AI Capacity

    Nvidia’s next Vera Rubin AI platform is expected to make it one of the first major deployments of HBM4, the new high-bandwidth stacked memory technology. SK hynix has reportedly secured about 70% of Nvidia’s HBM4 allocation for Rubin, according to Yonhap News Agency, a larger share than earlier...