VR2R-0070_r1

VR2R-0070

① SA Source

• Source: 開啟完整 SA 文章
• Section: Rubin
• Line hint: 63

Context Before

Notably, the Tensor core width doubling only applies to FP4 and FP8, with BF16 and TF32 remaining the same as Blackwell, resulting in performance scaling only 1.6x of Blackwell. This architectural decision reflects NVIDIA’s belief that most training and inference workloads will move away from TF32 and BF16 and onto FP8 and FP4.

Evidence

On the memory front, the move to HBM4 means double the bus width per stack, running at 10.8 GT/s for 22TB/s total bandwidth or 2.75x Blackwell at the same 288GB capacity as GB300.

Context After

Memory bandwidth has been upgraded significantly from the original 13TB/s advertised at GTC 2025. In order to catch up to AMD MI450’s memory bandwidth, Nvidia requested much higher HBM4 pin speeds from the DRAM suppliers - well above the speeds that was in the JEDEC specification for HBM4.

② Atomic Claim

在 memory 方面，轉向 HBM4 使每 stack bus width 加倍，以 10.8 GT/s 運作並提供 22TB/s total bandwidth，約為 Blackwell 的 2.75 倍；capacity 則與 GB300 相同，維持 288GB。

• Epistemic Mode: ASSERTED
• Mapping Status: COMPLETE

③ Semantic Frame

{
  "comparison_expression": "HBM4 bus width doubles; 10.8 GT/s yields 22TB/s total bandwidth, 2.75x Blackwell, while capacity remains 288GB as on GB300.",
  "entities": [
    {
      "id": "04_knowledge_base/HBM4",
      "label": "HBM4"
    },
    {
      "id": "04_knowledge_base/Blackwell",
      "label": "Blackwell"
    },
    {
      "id": "04_knowledge_base/GB300",
      "label": "GB300"
    }
  ],
  "frame_type": "COMPARISON",
  "metric": "HBM4_BUS_BANDWIDTH_AND_CAPACITY",
  "operator": "MIXED_SCALING",
  "qualifiers": {
    "condition_text": null,
    "numeric_mentions": [
      "2x",
      "10.8 GT/s",
      "22TB/s",
      "2.75x",
      "288GB"
    ],
    "temporal_mentions": []
  }
}

④ Canonical Entity Mapping

Role	Surface Label	Canonical Target
entity	HBM4	HBM4
entity	Blackwell	Blackwell
entity	GB300	GB300

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。