# 使用容器启动vllm，双卡运行，请求频繁会挂掉

**URL:** <https://discuss.vllm.ai/t/vllm/1221>\
**Category:** General\
**Created:** [July 29, 2025, 10:10am UTC](https://discuss.vllm.ai/t/vllm/1221 "2025-07-29T10:10:05Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![aproudcandefeat](https://yyz2.discourse-cdn.com/flex036/user_avatar/discuss.vllm.ai/aproudcandefeat/32/543_2.png) [@aproudcandefeat](https://discuss.vllm.ai/u/aproudcandefeat)\
**Post date:** [July 29, 2025, 10:10am UTC](https://discuss.vllm.ai/t/vllm/1221/1 "2025-07-29T10:10:05Z")

</div>

使用qwen3-14B，关掉了思考模式，使用两张A100 40G启动该服务。跑的时间久的时候会出现这个错误，降低gpu\_memory\_utilization会改善一点。

报错内容如下:  
500 Internal Server Error  
ERROR: Exception in ASGI application

vllm.v1.engine.exceptions.EngineDeadError: EngineCore encountered an issue. See stack trace (above) for the root cause.

---

<div class="post-metadata">

**Author:** ![RunLLM](https://yyz2.discourse-cdn.com/flex036/user_avatar/discuss.vllm.ai/runllm/32/233_2.png) [@RunLLM](https://discuss.vllm.ai/u/RunLLM)\
**Post date:** [July 29, 2025, 10:10am UTC](https://discuss.vllm.ai/t/vllm/1221/2 "2025-07-29T10:10:46Z")

</div>

根据报错内容，vLLM 在长时间运行后出现 500 Internal Server Error，日志显示 vllm.v1.engine.exceptions.EngineDeadError: EngineCore encountered an issue。你提到降低 gpu\_memory\_utilization 有所改善，但问题依然存在。

这种 EngineDeadError 通常是由于显存不足、CUDA 非法访问或资源耗尽等问题导致 EngineCore 崩溃。建议进一步降低 gpu\_memory\_utilization，或减少并发、上下文长度等参数，确保显存有足够冗余。此外，可参考官方文档的[故障排查建议](https://docs.vllm.ai/en/stable/getting_started/troubleshooting.html)，如开启 VLLM\_LOGGING\_LEVEL=DEBUG、CUDA\_LAUNCH\_BLOCKING=1 等环境变量以定位具体原因。

需要更详细的排查步骤吗？

Sources:

- [Troubleshooting — vLLM](https://docs.vllm.ai/en/stable/getting_started/troubleshooting.html)
- [vllm.v1.engine.exceptions.EngineDeadError](https://docs.vllm.ai/en/stable/api/vllm/vllm.v1.engine.exceptions.html)