VR2R-0137
① SA Source
- Source: 開啟完整 SA 文章
- Section:
BlueField-4 - Line hint:
113
Context Before
BlueField-4’s design departs significantly from BlueField-3.
Evidence
Instead of doing a bespoke tapeout with compute and networking, NVIDIA simply reuses their large Grace CPU die, co-packaged with a ConnectX-9 die to make an 800G DPU with massive compute capabilities.
Context After
128GB of LPDDR5 feeds the Grace CPU at half the bandwidth of regular Grace. That is 4x the memory capacity of BlueField-3. BlueField-4 can also function as a storage controller, with four BF-4 chips in each Context Memory Storage system.
② Atomic Claim
NVIDIA 不再做 compute/networking 的 bespoke tapeout,而是重用大型 Grace CPU die,與 ConnectX-9 die co-package,形成具備大量 compute capability 的 800G DPU。
- Epistemic Mode:
ASSERTED - Mapping Status:
COMPLETE
③ Semantic Frame
{
"frame_type": "NARY_RELATION",
"participants": [
{
"node": {
"id": "02_companies/NVDA",
"label": "NVIDIA"
},
"role": "vendor"
},
{
"node": {
"id": "04_knowledge_base/NVIDIA Grace",
"label": "Grace CPU"
},
"role": "cpu_die"
},
{
"node": {
"id": "04_knowledge_base/ConnectX-9",
"label": "ConnectX-9"
},
"role": "network_die"
},
{
"node": {
"id": "04_knowledge_base/DPU",
"label": "DPU"
},
"role": "product_type"
}
],
"qualifiers": {
"condition_text": null,
"numeric_mentions": [
"800G"
],
"temporal_mentions": []
},
"relation_type": "CO_PACKAGES_GRACE_AND_CONNECTX9_FOR_DPU"
}④ Canonical Entity Mapping
| Role | Surface Label | Canonical Target |
|---|---|---|
| vendor | NVIDIA | NVDA |
| cpu_die | Grace CPU | 04_knowledge_base/NVIDIA Grace |
| network_die | ConnectX-9 | ConnectX-9 |
| product_type | DPU | DPU |
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。