Unverified50% confidenceFactExact time
在35B参数模型中GEMM内核往往占据总计算时间的60-80%
1
Sources
50%
Confidence
Long-term
Relevance
7/20/2026
First Seen
Sources
Related Claims
UnverifiedH100内存子系统的动态功耗在大模型推理中占比可达整卡功耗的20%～35%72% similarUnverifiedThe 34B 80-layer model increases inference time by approximately 18% compared to the 27B69% similarUnverified参数量在1亿以下的轻量模型经领域数据微调后可达95%以上意图分类准确率，单次推理延迟控制在5-20ms以内66% similarUnverified业界顶级集群的MFU（模型算力利用率）通常在35%-60%之间65% similarUnverifiedIn a complete AI application system, GPU inference typically accounts for only 30%–40% of total compute, with the remaining 60%–70% of the compute load borne by the CPU.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/566988API
curl https://kongchang.com/api/v1/knowledge/claims/566988MCP
get_claim(id=566988)