clm_0b5a9fa1fd734a71bfb2627451855d31_r1

IX2-0519

① SA Source

• Source: 開啟完整 SA 文章
• Section: GPT-OSS 120B Single Node
• Line hint: 742

Context Before

GPT-OSS 120B Single Node

MI300X, MI325X, H200, and H100 group in the lower-left of the throughput vs interactivity plot, indicating broadly similar tradeoffs, with Nvidia generally holding a modest lead. The next step up is MI355X, which delivers roughly more than 2x higher token throughput per GPU at a given interactivity level, relative to that first group. Within MI355X, ATOM shifts the curve toward higher throughput at low interactivity, suggesting it prioritizes peak throughput over per-user responsiveness.

Evidence

While B200 and GB200 share the same Blackwell compute die, GB200 achieves a higher throughput–interactivity curve because the platform and serving stack reduce non-compute bottlenecks at scale (interconnect/topology, CPU-GPU coupling

Context After

Source: SemiAnalysis InferenceX ↗

② Atomic Claim

雖然 B200 與 GB200 採用相同的 Blackwell compute die，但 GB200 的 throughput–interactivity 曲線更高，因其平台與 serving stack 能在大規模部署時降低非運算瓶頸，包括 interconnect/topology 與 CPU-GPU coupling。

• Epistemic Mode: ASSERTED
• Mapping Status: COMPLETE

③ Semantic Frame

{
  "comparison_expression": "雖然 B200 與 GB200 採用相同的 Blackwell compute die，但 GB200 的 throughput–interactivity 曲線更高，因其平台與 serving stack 能在大規模部署時降低非運算瓶頸，包括 interconnect/topology 與 CPU-GPU coupling。",
  "entities": [
    {
      "id": "04_knowledge_base/NVIDIA B200",
      "label": "B200"
    },
    {
      "id": "04_knowledge_base/GB200",
      "label": "GB200"
    },
    {
      "id": "04_knowledge_base/Blackwell",
      "label": "Blackwell"
    },
    {
      "id": "04_knowledge_base/CPU",
      "label": "CPU"
    },
    {
      "id": "04_knowledge_base/GPU",
      "label": "GPU"
    }
  ],
  "frame_type": "COMPARISON",
  "metric": "COMPUTE_PERFORMANCE",
  "operator": "EQUAL_TO",
  "qualifiers": {
    "condition_text": null,
    "numeric_mentions": [],
    "temporal_mentions": []
  }
}

④ Canonical Entity Mapping

Role	Surface Label	Canonical Target
comparison_entity_0	B200	04_knowledge_base/NVIDIA B200
comparison_entity_1	GB200	GB200
comparison_entity_2	Blackwell	Blackwell
comparison_entity_3	CPU	CPU
comparison_entity_4	GPU	GPU

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。