clm_2606414c00604ef19ffa2f308101403c_r1

IX2-0416

① SA Source

• Source: 開啟完整 SA 文章
• Section: WideEP
• Line hint: 646

Context Before

The obvious way to scale is replication: deploy N independent EP8 instances across N nodes. Each instance serves requests independently with no cross-node communication. This scales throughput linearly, but each GPU still holds 32 experts per layer, and each token activates at most 8 of those 32 local experts. 75% of expert weights sit cold in HBM.

Wide expert parallelism (WideEP) takes a different approach by scaling EP _across _nodes rather than replicating independent instances. On a 64-GPU cluster (8 nodes), DP64/EP64 places only 256/64 = 4 experts per layer per GPU, each still holding a full replica of the non-expert weights. During the MoE phase, tokens from all 64 DP ranks are dispatched via all-to-all to the GPUs hosting their routed experts.

Evidence

First, reducing expert footprint from 32 to 4 experts/GPU frees substantial HBM for KV cache, directly increasing per-GPU batch size capacity

Context After

_A WideEP EP64 DP64 deployment of DeepSeek R1. All 256 experts per layer are divided evenly among the 64 GPUs (8 nodes), and attention and other non-expert weights (shared expert, gating network, RMSNorm, LM head, etc.) are replicated across all 64 DP ranks. _Source: SemiAnalysis

② Atomic Claim

第一，把每顆 GPU 的 expert footprint 從 32 個降到 4 個，可釋放大量 HBM 給 KV cache，直接提高每顆 GPU 可承載的 batch size。

• Epistemic Mode: ASSERTED
• Mapping Status: COMPLETE

③ Semantic Frame

{
  "attribute": "COUNT",
  "context_nodes": [
    {
      "id": "04_knowledge_base/HBM",
      "label": "HBM"
    },
    {
      "id": "04_knowledge_base/KV cache",
      "label": "KV cache"
    },
    {
      "id": "04_knowledge_base/Batch size",
      "label": "batch size"
    }
  ],
  "entity": {
    "id": "04_knowledge_base/GPU",
    "label": "GPU"
  },
  "frame_type": "ATTRIBUTE",
  "qualifiers": {
    "condition_text": null,
    "numeric_mentions": [
      "32",
      "4"
    ],
    "temporal_mentions": []
  },
  "value": {
    "numeric_mentions": [
      "32",
      "4"
    ],
    "value_text": "第一，把每顆 GPU 的 expert footprint 從 32 個降到 4 個，可釋放大量 HBM 給 KV cache，直接提高每顆 GPU 可承載的 batch size。"
  }
}

④ Canonical Entity Mapping

Role	Surface Label	Canonical Target
entity	GPU	GPU
context_0	HBM	HBM
context_1	KV cache	04_knowledge_base/KV cache
context_2	batch size	04_knowledge_base/Batch size

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。