Unverified50% confidenceTradeoffExact time
单个70B参数模型可能占用40GB以上显存，单张A100仅能承载极有限的并发请求
1
Sources
50%
Confidence
Long-term
Relevance
7/11/2026
First Seen
Sources
Ollama获6500万美元B轮：85%财富500强已部署本地大模型
redditr/ollama7/10/2026
Related Claims
Unverified70B级别的模型即便经过量化，往往也需要40GB以上显存或多卡配置77% similarUnverified支持128K上下文的70B参数模型的KV Cache满负荷时可能需要数百GB显存，超过单张H100的80GB显存容量77% similarUnverified1750亿参数模型仅模型权重（BF16精度）就需350GB存储，Adam优化器需维护一阶矩和二阶矩两份状态，合计超过1TB71% similarUnverified量化格式可将70B参数模型的运行内存从140GB压缩至40GB以内71% similarUnverified对于数十亿参数的SD模型，LoRA文件通常只有10-200MB70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/486502API
curl https://kongchang.com/api/v1/knowledge/claims/486502MCP
get_claim(id=486502)