SA260324R-0295_r1

SA260324R-0295

① SA Source

• Source: 開啟完整 SA 文章
• Section: CMX
• Line hint: 374

Context Before

KV cache grows linearly with input sequence length and number of users and is the primary tradeoff when it comes to prefill performance (time to first token). At scale, on-device HBM does not have enough capacity.

Evidence

Host DRAM extends beyond HBM capacity with an additional tier of cache, but also hits limits on total amount per node, memory bandwidth, and network bandwidth.

Context After

Enter NVMe storage for additional KVcache offload.

② Atomic Claim

Host DRAM 可在 HBM 之外增加一層 cache，但同樣會受到每 node 總容量、memory bandwidth 與 network bandwidth 限制。

• Epistemic Mode: ASSERTED
• Mapping Status: COMPLETE

③ Semantic Frame

{
  "frame_type": "NARY_RELATION",
  "participants": [
    {
      "node": {
        "id": "04_knowledge_base/DRAM",
        "label": "DRAM"
      },
      "role": "host_memory"
    },
    {
      "node": {
        "id": "04_knowledge_base/HBM",
        "label": "HBM"
      },
      "role": "device_memory"
    }
  ],
  "qualifiers": {
    "condition_text": "limits include total amount per node, memory bandwidth, and network bandwidth",
    "numeric_mentions": [],
    "temporal_mentions": []
  },
  "relation_type": "HOST_DRAM_EXTENDS_HBM_WITH_LIMITS"
}

④ Canonical Entity Mapping

Role	Surface Label	Canonical Target
host_memory	DRAM	DRAM
device_memory	HBM	HBM

⑤ Human Review

請在 Properties 逐項確認：

• 原文 → Atomic Claim 是否忠實
• Atomic Claim → Semantic Frame 是否忠實
• Canonical Entity mapping 是否正確
• Epistemic mode 是否保留原文語氣
• 最後選擇 review_action

Review state

Markdown 內文不是正式 approval。只有 Apply bridge 寫入的 Decision Ledger event 才是正式決策。