2025-06-13_amd-advancing-ai-mi350x-and-mi400-ualoe72-mi500-ual256::AMD25-0005
① SA Source
- Source: 開啟完整 SA 文章
- Section:
Executive Summary - Line hint:
37
Context Before
Despite AMD’s marketing RDF ↗, the MI355 128 GPU rack is not a “rack scale solution” – it only has a scale up world size of 8 GPUs versus the GB200 NVL72 which has a world size of 72 GPUs. The GB200 NVL72 will beat the MI355X on Perf per TCO for large frontier reasoning model inference
Evidence
The MI355X will have similar collective performance as the HGX B200 but MI355X collectives will run at least 18x slower than on the GB200 NVL72, if not even slower
Context After
AMD announced its Developer Cloud, which will bring on demand pricing to 3.00/hr/GPU in the current AMD Neocloud market in a move that could potentially make renting AMD GPUs competitive vs renting Nvidia GPUs
② Atomic Claim
SemiAnalysis 預期 MI355X collective performance 約等於 HGX B200，但至少比 GB200 NVL72 慢 18 倍。
- Epistemic Mode:
EXPECTED - Mapping Status:
COMPLETE
③ Semantic Frame
{
"comparison_expression": "MI355X ~HGX B200; >=18x slower than GB200 NVL72",
"entities": [
{
"id": "04_knowledge_base/MI355X",
"label": "MI355X"
},
{
"id": "04_knowledge_base/NVIDIA B200",
"label": "B200"
},
{
"id": "04_knowledge_base/GB200 NVL72",
"label": "GB200 NVL72"
}
],
"frame_type": "COMPARISON",
"metric": "collective_performance",
"operator": "AT_LEAST_18X_SLOWER_THAN_GB200",
"qualifiers": {
"condition_text": null,
"numeric_mentions": [
"18x"
],
"temporal_mentions": []
}
}④ Canonical Entity Mapping
| Role | Surface Label | Canonical Target |
|---|---|---|
| comparison_entity_0 | MI355X | MI355X |
| comparison_entity_1 | B200 | 04_knowledge_base/NVIDIA B200 |
| comparison_entity_2 | GB200 NVL72 | 04_knowledge_base/GB200 NVL72 |
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。