VR2R-0069_r1

VR2R-0069

① SA Source

• Source: 開啟完整 SA 文章
• Section: Rubin
• Line hint: 61

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

Evidence

inference workloads will move away from TF32 and BF16 and onto FP8 and FP4.

Context After

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. 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

……inference workloads 將逐步離開 TF32 與 BF16，轉向 FP8 與 FP4。

• Epistemic Mode: ASSERTED
• Mapping Status: PARTIAL

③ Semantic Frame

{
  "comparison_expression": "inference workloads 將由 TF32/BF16 轉向 FP8/FP4。",
  "entities": [
    {
      "id": "04_knowledge_base/TF32",
      "label": "TF32"
    },
    {
      "id": "04_knowledge_base/BF16",
      "label": "BF16"
    },
    {
      "id": "04_knowledge_base/FP8",
      "label": "FP8"
    },
    {
      "id": "04_knowledge_base/FP4",
      "label": "FP4"
    }
  ],
  "frame_type": "COMPARISON",
  "metric": "WORKLOAD_PRECISION_DIRECTION",
  "operator": "SHIFTS_FROM_TO",
  "qualifiers": {
    "condition_text": "承接 VR2R-0068；原句主詞與 training 部分被切在前一 claim",
    "numeric_mentions": [],
    "temporal_mentions": []
  }
}

④ Canonical Entity Mapping

Role	Surface Label	Canonical Target
entity	TF32	TF32
entity	BF16	BF16
entity	FP8	FP8
entity	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。