clm_d668b74367aa4636866ce9a7946192dd_r1

IX2-0259

① SA Source

• Source: 開啟完整 SA 文章
• Section: Nvidia Blackwell Perf TCO Analysis - B100 vs B200 vs GB200NVL72
• Line hint: 430

Context Before

Source: SemiAnalysis InferenceX ↗

Evidence

So when we need more than 8 GPUs (but less than 72 GPUs) for the inference setup, we need to bring in multiple nodes of B300 servers to form our inference system which means communications falls back to the lower InfiniBand XDR scale-out network featuring 800Gbit/s (uni-di) per GPU of bandwidth

Context After

SemiAnalysis is free open source software and reader-supported. To receive new posts and support our work, consider becoming a free or paid subscriber.

Subscribed

② Atomic Claim

因此當推論配置需要超過 8 顆 GPUs、但少於 72 顆 GPUs 時，B300 必須使用多個 nodes 組成系統，通訊就會退回頻寬較低的 InfiniBand XDR scale-out network，每顆 GPU 僅有 800Gbit/s uni-di。

• Epistemic Mode: ASSERTED
• Mapping Status: PARTIAL

③ Semantic Frame

{
  "comparison_expression": "因此當推論配置需要超過 8 顆 GPUs、但少於 72 顆 GPUs 時，B300 必須使用多個 nodes 組成系統，通訊就會退回頻寬較低的 InfiniBand XDR scale-out network，每顆 GPU 僅有 800Gbit/s uni-di。",
  "entities": [
    {
      "id": "04_knowledge_base/GPU",
      "label": "GPUs"
    },
    {
      "id": "04_knowledge_base/NVIDIA B300",
      "label": "B300"
    },
    {
      "id": "04_knowledge_base/InfiniBand",
      "label": "InfiniBand"
    },
    {
      "id": "04_knowledge_base/Scale-out networking",
      "label": "scale-out network"
    }
  ],
  "frame_type": "COMPARISON",
  "metric": "BANDWIDTH",
  "operator": "LESS_THAN",
  "qualifiers": {
    "condition_text": null,
    "numeric_mentions": [
      "8",
      "72",
      "800Gb"
    ],
    "temporal_mentions": []
  }
}

④ Canonical Entity Mapping

Role	Surface Label	Canonical Target
comparison_entity_0	GPUs	GPU
comparison_entity_1	B300	04_knowledge_base/NVIDIA B300
comparison_entity_2	InfiniBand	InfiniBand
comparison_entity_3	scale-out network	04_knowledge_base/Scale-out networking

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。