8 news tagged with 008060 in the last 7 days
Original: SOCAMM2 용량 반토막…엔비디아 고육지책에 기판 업계는 '방긋'
Nvidia is cutting SOCAMM2 memory capacity from 192GB to 96GB per module (LPDDR5X stack 4-high→2-high) to work around LPDDR shortages for its Vera CPU-based AI servers. Module and MCP substrate counts stay the same (8 modules + 32 substrates per Vera CPU), so Korean substrate suppliers (Simmtech, Daeduck Electronics, Korea Circuit) are net winners as more servers can be built off the same LPDDR pool. Memory players (Samsung, SK Hynix) read the cut as proof LPDDR demand from AI servers + RTX Spark AI PCs is outrunning supply, not weakening.
Why it matters: TheElec scoop on a concrete Nvidia spec change with clearly named beneficiaries — Korean substrate makers gain unit volume while LPDDR shortage signal is positive for memory names.
Open source article