2025-12-04_aws-trainium3-deep-dive-a-potential::TRN3-0025

① SA Source

Context Before

image

Source: SemiAnalysis

Evidence

The rest of the items in the compute tray, the PCIe 6.0 x16 retimers, Nitro-v6 NICs, PCIe 6.0 x16 AEC cage, DIMMs, and 2x8TB local NVMe drives – are air-cooled using fans.

Context After

Turning to the host CPU, Graviton4 will be the only time to market CPU option for Trainium3 NL72x2 Switched. The CPU can be upgraded later to the next generation of Graviton during the Trainium3’s lifecycle. In theory, x86 CPUs are also supported as these can also interface with other components via PCIe, but we don’t believe that they will plan to have an x86 Trainium3 NL72x2 Switched SKU and will only offer an x86 Trainium NL32x2 Switched SKU. Because Trainium3 uses PCIe 6.0 and Graviton4 uses PCIe 5.0, two PCIe gearboxes must be placed next to the CPU to convert from PCIe 6.0 to PCIE 5.0 for communication between the CPU and GPUs. For CPU memory, 12 DDR5 DIMM slots are placed next to the CPU with DDR5 DIMM modules of 64GB and 128GB capacities to be used for the mainstream SKU. Two 8TB local NVMe drives per compute tray will be used for local storage.

Trainium3 NL72x2 Switched Scale Out Networking

② Atomic Claim

NL72x2 compute tray 的 retimers、Nitro-v6 NIC、AEC cage、DIMMs 與 2×8TB NVMe 等其餘元件以 fans air-cooling。

  • Epistemic Mode: ASSERTED
  • Mapping Status: COMPLETE

③ Semantic Frame

{
  "frame_type": "NARY_RELATION",
  "participants": [
    {
      "node": {
        "id": "04_knowledge_base/Compute Tray",
        "label": "Compute Tray"
      },
      "role": "tray"
    },
    {
      "node": {
        "id": "04_knowledge_base/PCIe",
        "label": "PCIe"
      },
      "role": "pcie"
    },
    {
      "node": {
        "id": "04_knowledge_base/Nitro",
        "label": "Nitro"
      },
      "role": "nic"
    },
    {
      "node": {
        "id": "04_knowledge_base/NVMe",
        "label": "NVMe"
      },
      "role": "storage"
    }
  ],
  "qualifiers": {
    "condition_text": "NL72x2",
    "numeric_mentions": [
      "2x8TB"
    ],
    "temporal_mentions": []
  },
  "relation_type": "AIR_COOLED_COMPONENTS"
}

④ Canonical Entity Mapping

RoleSurface LabelCanonical Target
trayCompute Tray04_knowledge_base/Compute Tray
pciePCIePCIe
nicNitroNitro
storageNVMeNVMe

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。