NIEK2-0106
① SA Source
- Source: 開啟完整 SA 文章
- Section:
GPU and LPU Integration: Attention FFN Disaggregation (AFD) - Line hint:
93
Context Before
Source: SemiAnalysis
Evidence
During decode phase, the GPU utilization of attention barely improves when scaling batch size due to being bounded by loading KV cache
Context After
This is something we have worked with certain hardware vendors and memory companies on with our inference simulator for more than 6 months. ↗
② Atomic Claim
在 decode 階段，attention 因受到 KV cache 載入限制，即使提高 batch size，GPU 利用率也幾乎不會改善。
- Epistemic Mode:
ASSERTED - Mapping Status:
COMPLETE
③ Semantic Frame
{
"attribute": "UTILIZATION",
"context_nodes": [
{
"id": "04_knowledge_base/KV cache",
"label": "KV cache"
},
{
"id": "04_knowledge_base/Batch size",
"label": "batch size"
},
{
"id": "04_knowledge_base/GPU",
"label": "GPU"
}
],
"entity": {
"id": "04_knowledge_base/Decode",
"label": "decode"
},
"frame_type": "ATTRIBUTE",
"qualifiers": {
"condition_text": null,
"numeric_mentions": [],
"temporal_mentions": []
},
"value": {
"numeric_mentions": [],
"value_text": "在 decode 階段，attention 因受到 KV cache 載入限制，即使提高 batch size，GPU 利用率也幾乎不會改善。"
}
}④ Canonical Entity Mapping
| Role | Surface Label | Canonical Target |
|---|---|---|
| entity | decode | Decode |
| context_0 | KV cache | 04_knowledge_base/KV cache |
| context_1 | batch size | 04_knowledge_base/Batch size |
| context_2 | 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。