Topic
Explore English articles about Memory & Storage.
zHBM, unveiled by Samsung at FMS 2026, is a new concept that physically shortens data travel distance by moving HBM from beside the processor to above it. Using HBM5's projected bandwidth of roughly 4 TB/s as the baseline, Samsung claims an 8x improvement—but since HBM5's own specification is not yet finalized, verification of these figures will have to wait for mass production.
The NVMe 2.4 specification builds virtualization and live migration for locally attached SSDs directly into the standard. It shifts the complexity of storage virtualization—long borne by hypervisors—onto the SSD controller itself, opening a path to resolving the operational constraints of storage-optimized VMs.
Kioxia and Sandisk achieved an areal density of 37 Gb/mm² with 332-layer QLC 3D NAND flash, overtaking Samsung's V10 (28 Gb/mm²), which leads in layer count, in terms of density. Through the CBA architecture, Toggle DDR6.0, and PI-LTT technology, they simultaneously improved density, speed, and power efficiency, delivering a design that meets the large-capacity storage demands of AI data centers.
Versity has added RDMA support to its open-source S3 gateway, enabling direct transfers from tape archives to GPU memory. Amid soaring NVMe prices and server shortages, this low-cost setup aims for 20GBps using 50 parallel drives—how will it reshape storage design for AI training?
NVIDIA has open-sourced cuFile, its API for direct GPU storage access, and is driving standardization of the SCADA architecture—where GPUs control storage I/O without going through the CPU—together with more than 40 companies. As the bottleneck in AI inference shifts from compute to data delivery, structural change is beginning across the storage industry.
Lexar has listed a 32GB DDR5-7200 kit using domestic Chinese DRAM for 3,999 yuan on JD.com. While clock speeds have advanced, the emergence of a new supplier doesn't necessarily translate into cheaper memory.
The first OCP specification standardizes HBF's host connections, reliability, and software operations. After product samples ship, verification of interoperability and power/durability characteristics will continue on actual hardware running alongside HBM.
Global memory shortages are causing significant delays in MacBook Air delivery times, and even after June's price increase, balancing supply and demand remains difficult. Expanding AI-related demand is squeezing the supply of DRAM for PCs, and while Apple is exploring stockpiling and diversifying its sourcing, the shortage has yet to be resolved.
Montage has moved its CXL 3.2-compliant memory expansion chip into trial production, and Samsung and SK hynix have also completed initial validation. Attention now shifts to the mass-production timeline and real-world workload performance.
Reports indicate CXMT's LPDDR6, designed at 12.8Gbps, is nearing completion of development verification. Cross-referencing official disclosures with competitors reveals the customer evaluation and certification steps still required before mass production.
