SA260324R-0100_r1

SA260324R-0100

① SA Source

• Source: 開啟完整 SA 文章
• Section: GPU and LPU Integration: Attention FFN Disaggregation (AFD)
• Line hint: 105

Context Before

As we explained in our InferenceX article, LLM inference involves two phases: prefill and decode. Prefill processes the full input context: It is compute-intensive, which is suitable for GPUs. On the other hand, decode predicts new tokens and is memory-bounded. Decode is latency-sensitive because the model predicts new tokens one by one, and

Evidence

LPU’s high SRAM bandwidth and low-latency capabilities can help accelerate this iterative process.

Context After

Attention and FFN are subsets of operations in a model. In a model forward pass, attention’s output feeds into a token router, and the token router assigns each token to k experts, where each expert is an FFN. Attention and FFN have very different performance properties. During decode phase, the GPU utilization of attention barely improves when scaling batch size due to being bounded by loading KV cache. In contrast, the GPU utilization of FFN scales with batch size comparatively better.

② Atomic Claim

LPU 的高 SRAM 頻寬與低延遲能力，有助於加速這個反覆迭代的過程。

• Epistemic Mode: ASSERTED
• Mapping Status: PARTIAL

③ Semantic Frame

{
  "attribute": "PERFORMANCE_CHARACTERISTICS",
  "context_nodes": [
    {
      "id": "04_knowledge_base/SRAM",
      "label": "SRAM"
    }
  ],
  "entity": {
    "id": "04_knowledge_base/Groq LPU",
    "label": "LPU"
  },
  "frame_type": "ATTRIBUTE",
  "qualifiers": {
    "condition_text": "this iterative process 的指涉依賴前文",
    "numeric_mentions": [],
    "temporal_mentions": []
  },
  "value": {
    "numeric_mentions": [],
    "value_text": "高 SRAM 頻寬與低延遲能力可幫助加速原文所指的反覆迭代過程。"
  }
}

④ Canonical Entity Mapping

Role	Surface Label	Canonical Target
entity	LPU	04_knowledge_base/Groq LPU
context_node	SRAM	SRAM

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。