clm_4d775eaa2c1d453fb8ff56bdaebc5d16_r1

VR2-0178

① SA Source

• Source: 開啟完整 SA 文章
• Section: CPX Form Factor
• Line hint: 245

Context Before

GDDR7 offers materially lower cost per GB and avoids the need for 2.5D packaging,

Evidence

However, Nvidia began exploring HBM-equipped variants for prefill, either via modified CPX configurations or through lower memory spec (such as using HBM3E) Rubin deployments dedicated to prefill, which we noted ↗ this way back in early December last year ↗ in our Accelerator & HBM model ↗.

Context After

We also think a lot this shift is driven by evolving memory economics. **Conventional DRAM pricing has risen sharply: **As DDR pricing increases, the relative premium of HBM compresses because pricing is more locked down in long term contracts, narrowing the cost gap between a GDDR-based CPX and lower-spec HBM configurations, therefore eliminating a lot of the cost benefits GDDR offers relative to performance. While memory bandwidth is not as important for pre-fill compare to decode, it is still necessary.

Compute Tray Redesign

② Atomic Claim

不過，Nvidia 開始探索搭載 HBM 的 prefill 方案，包括修改後的 CPX 配置，或採較低 memory spec（例如 HBM3E）並專用於 prefill 的 Rubin；SemiAnalysis 早在去年 12 月初的 Accelerator 與 HBM model 已提及此方向。

• Epistemic Mode: ASSERTED
• Mapping Status: PARTIAL

③ Semantic Frame

{
  "comparison_expression": "不過，Nvidia 開始探索搭載 HBM 的 prefill 方案，包括修改後的 CPX 配置，或採較低 memory spec（例如 HBM3E）並專用於 prefill 的 Rubin；SemiAnalysis 早在去年 12 月初的 Accelerator 與 HBM model 已提及此方向。",
  "entities": [
    {
      "id": "02_companies/NVDA",
      "label": "Nvidia"
    },
    {
      "id": "04_knowledge_base/HBM",
      "label": "HBM"
    },
    {
      "id": "04_knowledge_base/Prefill",
      "label": "prefill"
    },
    {
      "id": "04_knowledge_base/CPX High-Density Interconnect Paradigm",
      "label": "CPX"
    },
    {
      "id": "04_knowledge_base/HBM3E",
      "label": "HBM3E"
    },
    {
      "id": "04_knowledge_base/Vera Rubin",
      "label": "Rubin"
    }
  ],
  "frame_type": "COMPARISON",
  "metric": "UNSPECIFIED_METRIC",
  "operator": "LESS_THAN",
  "qualifiers": {
    "condition_text": null,
    "numeric_mentions": [
      "12"
    ],
    "temporal_mentions": []
  }
}

④ Canonical Entity Mapping

Role	Surface Label	Canonical Target
comparison_entity_0	Nvidia	NVDA
comparison_entity_1	HBM	HBM
comparison_entity_2	prefill	Prefill
comparison_entity_3	CPX	04_knowledge_base/CPX High-Density Interconnect Paradigm
comparison_entity_4	HBM3E	HBM3E
comparison_entity_5	Rubin	04_knowledge_base/Vera Rubin

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。