Unverified50% confidenceFactExact time
在密集型中等规模模型（如27B）上AMD R9700 明显领先，MoE 模型上两者基本持平，小型负载上 B70 略占上风
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/5/2026
First Seen
Valid until: 1/3/2027
Sources
Related Entities
Related Claims
UnverifiedQwen3.5-122B-A10B 是总参数量约 122B、激活参数约 10B 的 MoE 架构模型64% similarUnverified部署DeepSeek-R1 671B MoE模型需要至少数张高端GPU（如H100或A100），显存需求超过320GB64% similarUnverified通过张量并行，2×H100可流畅运行FP16精度的70B模型并支持较高并发62% similarUnverified一个经过良好微调的27B-32B模型运行成本仅为400B+级别模型的十分之一62% similarUnverified选择迷你主机时优先看CPU型号后缀：AMD带U后缀（如7840U）为低功耗15-28W，H后缀（如7840HS）为45W性能释放版；同代同数字H版性能比U版高20-30%，追求性能选H62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/977980API
curl https://kongchang.com/api/v1/knowledge/claims/977980MCP
get_claim(id=977980)