Unverified50% confidenceBenchmarkExact time
在配备6核锐龙和64GB内存的RTX 5070上，Strata短对话约90 Token/秒，12.8万Token上下文约67 Token/秒，读取提示词约1300 Token/秒
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/5/2026
First Seen
Valid until: 1/3/2027
Sources
Related Entities
Related Claims
Verified在RTX 5070（12GB显存）配置下，本地Qwen3 9B模型的输出速度达到约70-76 token/秒71% similarUnverified旧款RTX 3090因24GB显存能容纳更多专家，预估运行Strata性能反而达100到140 Token/秒，说明此工作负载下显存容量胜过纯浮点算力71% similarUnverified在RTX 5060 Ti 16GB + 64GB内存环境下，Qwen3 27B生成响应式个人网站耗时1小时10分钟，平均每秒生成6.42个Token70% similarUnverified提示处理速度上RTX 4090约3000 tokens/s，M5 Max约800 tokens/s，英伟达在长文档/代码库输入场景优势明显70% similarVerified9600波特率下UART每秒可传输约960个字节（每字节需10个比特：1起始位+8数据位+1停止位）70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/978620API
curl https://kongchang.com/api/v1/knowledge/claims/978620MCP
get_claim(id=978620)