clm_84fa0a9c4f2a47d7880f4b3f4ae42009_r1

IX2-0141

① SA Source

• Source: 開啟完整 SA 文章
• Section: DeepSeek Disagg +WideEP Results Deep Dive
• Line hint: 222

Context Before

Source: SemiAnalysis InferenceX ↗

Evidence

When composing disagg prefill+wideEP with FP4 on the MI355X, we observe suffers subpar performance

Context After

Although theoretical modeling shows that disagg inference on MI355Xs should perform way better than single node, disagg actually performs worse for higher interactivity levels due to a lack of kernel and collective optimization in the ROCm software stack when composing multiple SOTA inference optimizations together.

② Atomic Claim

在 MI355X 上同時組合 disagg prefill+wideEP 與 FP4 時，實際效能明顯不佳。

• Epistemic Mode: ASSERTED
• Mapping Status: COMPLETE

③ Semantic Frame

{
  "attribute": "COMPOSABILITY",
  "context_nodes": [
    {
      "id": "04_knowledge_base/Prefill",
      "label": "prefill"
    },
    {
      "id": "04_knowledge_base/FP4",
      "label": "FP4"
    }
  ],
  "entity": {
    "id": "04_knowledge_base/MI355X",
    "label": "MI355X"
  },
  "frame_type": "ATTRIBUTE",
  "qualifiers": {
    "condition_text": null,
    "numeric_mentions": [],
    "temporal_mentions": []
  },
  "value": {
    "numeric_mentions": [],
    "value_text": "在 MI355X 上同時組合 disagg prefill+wideEP 與 FP4 時，實際效能明顯不佳。"
  }
}

④ Canonical Entity Mapping

Role	Surface Label	Canonical Target
entity	MI355X	MI355X
context_0	prefill	Prefill
context_1	FP4	FP4

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。