华为OceanStor M900推出PB级共享KV缓存，用于AI超算集群
Huawei OceanStor M900 Brings PB-Class Shared KV Cache to AI SuperPoDs
精选理由
华为新出的OceanStor M900存储，专门给AI超算集群用的，能提供PB级的共享KV缓存，延迟和带宽都很好。
华为OceanStor M900 AI存储针对超大规模推理，其Lingqu池式KV缓存每集群可达64PB，NPU到SSD延迟约60微秒，聚合带宽约40TB/s，KV-Aware调度声称性能达24 DWPD。
原文 · pandaily
Huawei OceanStor M900 Brings PB-Class Shared KV Cache to AI SuperPoDs
Huawei's OceanStor M900 AI memory storage targets hyperscale inference with Lingqu-pooled KV Cache up to 64 PB per cluster, ~60μs NPU-to-SSD latency, ~40 TB/s aggregate bandwidth, and KV-Aware scheduling claiming up to 24 DWPD.
