🛰️ Daily AI Frontier
‹ back to 2026-08-24

为什么手机内存进入英伟达机柜后,贵过HBM?

Industry & News Efficiency & Systems

Ranking

Overall 78
Content 90
Popularity N/A

No observed public metrics; popularity remains neutral/archived.

Representative image for 为什么手机内存进入英伟达机柜后,贵过HBM?

Merged summary

TL;DR - Nvidia’s next-generation AI racks are adopting enormous quantities of modular LPDDR5X memory for higher CPU bandwidth and lower power consumption, making their total LPDDR5X cost potentially exceed HBM. Tight supply may force Nvidia to halve Vera Rubin’s planned CPU-memory capacity.

  • The projected Kyber rack uses 216 TB of LPDDR5X costing nearly $2.8 million, versus 124.4 TB of HBM4E costing about $2.5 million.
  • SOCAMM2 packages LPDDR5X as replaceable server modules, offering substantially higher bandwidth, lower power, and a smaller footprint than DDR5 RDIMMs.
  • Agent workloads, CPU–GPU coherent memory, and growing data-processing demands are driving CPU memory from 480 GB on Grace to as much as 1.5 TB on Vera.
  • Nvidia may receive only about 60% of its projected LPDRAM needs; reported mitigations include reducing Vera CPU memory from 1.5 TB to 768 GB, with shortages potentially lasting through 2027.

Sources (1)

为什么手机内存进入英伟达机柜后,贵过HBM?

雷峰网 (AI科技评论) 2026-08-24
Public signals N/A
Providers: Hugging Face · N/A OpenAlex · N/A Publisher · N/A Semantic Scholar · N/A X · N/A Fetched 2026-09-23 14:19:24.656997 UTC

TL;DR - Nvidia’s next-generation AI racks are adopting enormous quantities of modular LPDDR5X memory for higher CPU bandwidth and lower power consumption, making their total LPDDR5X cost potentially exceed HBM. Tight supply may force Nvidia to halve Vera Rubin’s planned CPU-memory capacity.

  • The projected Kyber rack uses 216 TB of LPDDR5X costing nearly $2.8 million, versus 124.4 TB of HBM4E costing about $2.5 million.
  • SOCAMM2 packages LPDDR5X as replaceable server modules, offering substantially higher bandwidth, lower power, and a smaller footprint than DDR5 RDIMMs.
  • Agent workloads, CPU–GPU coherent memory, and growing data-processing demands are driving CPU memory from 480 GB on Grace to as much as 1.5 TB on Vera.
  • Nvidia may receive only about 60% of its projected LPDRAM needs; reported mitigations include reducing Vera CPU memory from 1.5 TB to 768 GB, with shortages potentially lasting through 2027.
item →