Revati让大模型服务无需GPU即可快速模拟，精度超95%。
Revati: Transparent GPU-Free Time-Warp Emulation for LLM Serving
- 直接运行真实服务代码，通过时间跳跃模拟GPU计算耗时。
- 在多模型和并行配置下误差低于5%，速度比实机快5-17倍。
- 适合需要高效测试多种部署配置的研究者与工程师。
高效部署大模型需测试数百种服务配置，但实际GPU集群评估耗时数小时且成本高达数千美元。传统离散事件模拟器虽更快更便宜，却需重写服务系统的控制逻辑，随着框架迭代负担加重。本文提出Revati，一种透明的无GPU时间跃迁模拟器，可直接执行真实服务系统代码，达到近似仿真速度。该系统拦截CUDA API调用，虚拟化设备管理，使服务框架无需物理GPU即可运行；不真正执行GPU内核，而是根据预测的内核耗时进行时间跳跃。我们设计了一种协调协议，确保分布式进程间时间跳跃同步并保持因果关系。在vLLM和SGLang上，Revati在多个模型和并行配置下预测误差小于5%，运行速度比真实GPU执行快5-17倍。
原文摘要 · Abstract (English)
Deploying LLMs efficiently requires testing hundreds of serving configurations, but evaluating each one on a GPU cluster takes hours and costs thousands of dollars. Discrete-event simulators are faster and cheaper, but they require re-implementing the serving system's control logic -- a burden that compounds as frameworks evolve. We present Revati, a time-warp emulator that enables performance modeling by directly executing real serving system code at simulation-like speed. The system intercepts CUDA API calls to virtualize device management, allowing serving frameworks to run without physical GPUs. Instead of executing GPU kernels, it performs time jumps -- fast-forwarding virtual time by predicted kernel durations. We propose a coordination protocol that synchronizes these jumps across distributed processes while preserving causality. On vLLM and SGLang, Revati achieves less than 5% prediction error across multiple models and parallelism configurations, while running 5-17x faster than real GPU execution.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。