# RuntimeError: CUDA driver initialization failed, you might not have a CUDA gpu

**URL:** <https://discuss.vllm.ai/t/runtimeerror-cuda-driver-initialization-failed-you-might-not-have-a-cuda-gpu/2698>\
**Category:** General\
**Created:** [June 5, 2026, 12:52am UTC](https://discuss.vllm.ai/t/runtimeerror-cuda-driver-initialization-failed-you-might-not-have-a-cuda-gpu/2698 "2026-06-05T00:52:10Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![wangxuasp](https://yyz2.discourse-cdn.com/flex036/user_avatar/discuss.vllm.ai/wangxuasp/32/1453_2.png) [@wangxuasp](https://discuss.vllm.ai/u/wangxuasp)\
**Post date:** [June 5, 2026, 12:52am UTC](https://discuss.vllm.ai/t/runtimeerror-cuda-driver-initialization-failed-you-might-not-have-a-cuda-gpu/2698/1 "2026-06-05T00:52:10Z")

</div>

hardware： CPU：Hygon 7380 + GPU： Nvidia H800\*3 + memory：512GB

software：Ubuntu 24+torch==2.11.0+cu130 + vllm==0.21.0

driver：nvidia-smi==Driverversion: 580.159.04 CUDA Version: 13.0

startup log：(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] WorkerProc failed to start.  
(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] Traceback (most recent call last):  
(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/executor/multiproc\_executor.py”, line 837, in worker\_main  
(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] worker = WorkerProc(\*args, \*\*kwargs)  
(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] ^^^^^^^^^^^^^^^^^^^^^^^^^^^  
(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/tracing/otel.py”, line 178, in sync\_wrapper  
(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] return func(\*args, \*\*kwargs)  
(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] ^^^^^^^^^^^^^^^^^^^^^  
(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/executor/multiproc\_executor.py”, line 611, in **init**  
(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] self.worker.init\_device()  
(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/worker/worker\_base.py”, line 317, in init\_device  
(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] self.worker.init\_device() # type: ignore  
(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] ^^^^^^^^^^^^^^^^^^^^^^^^^  
(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/tracing/otel.py”, line 178, in sync\_wrapper  
(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] return func(\*args, \*\*kwargs)  
(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] ^^^^^^^^^^^^^^^^^^^^^  
(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/worker/gpu\_worker.py”, line 275, in init\_device  
(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] torch.accelerator.set\_device\_index(self.device)  
(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/torch/accelerator/ **init**.py”, line 191, in set\_device\_index  
(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] torch.\_C.\_accelerator\_setDeviceIndex(device\_index)  
(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/torch/cuda/ **init**.py”, line 478, in \_lazy\_init  
(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] torch.\_C.\_cuda\_init()  
(Worker pid=122743) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] RuntimeError: CUDA driver initialization failed, you might not have a CUDA gpu.  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] WorkerProc failed to start.  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] Traceback (most recent call last):  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/executor/multiproc\_executor.py”, line 837, in worker\_main  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] worker = WorkerProc(\*args, \*\*kwargs)  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] ^^^^^^^^^^^^^^^^^^^^^^^^^^^  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/tracing/otel.py”, line 178, in sync\_wrapper  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] return func(\*args, \*\*kwargs)  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] ^^^^^^^^^^^^^^^^^^^^^  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/executor/multiproc\_executor.py”, line 611, in **init**  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] self.worker.init\_device()  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/worker/worker\_base.py”, line 317, in init\_device  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] self.worker.init\_device() # type: ignore  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] ^^^^^^^^^^^^^^^^^^^^^^^^^  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/tracing/otel.py”, line 178, in sync\_wrapper  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] return func(\*args, \*\*kwargs)  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] ^^^^^^^^^^^^^^^^^^^^^  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/worker/gpu\_worker.py”, line 275, in init\_device  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] torch.accelerator.set\_device\_index(self.device)  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/torch/accelerator/ **init**.py”, line 191, in set\_device\_index  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] torch.\_C.\_accelerator\_setDeviceIndex(device\_index)  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/torch/cuda/ **init**.py”, line 478, in \_lazy\_init  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] torch.\_C.\_cuda\_init()  
(Worker pid=122742) ERROR 05-31 14:55:08 [multiproc\_executor.py:870] RuntimeError: CUDA driver initialization failed, you might not have a CUDA gpu.  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] EngineCore failed to start.  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] Traceback (most recent call last):  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/engine/core.py”, line 1114, in run\_engine\_core  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] engine\_core = EngineCoreProc(\*args, engine\_index=dp\_rank, \*\*kwargs)  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/tracing/otel.py”, line 178, in sync\_wrapper  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] return func(\*args, \*\*kwargs)  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] ^^^^^^^^^^^^^^^^^^^^^  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/engine/core.py”, line 880, in **init**  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] super(). **init** (  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/engine/core.py”, line 118, in **init**  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] self.model\_executor = executor\_class(vllm\_config)  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] ^^^^^^^^^^^^^^^^^^^^^^^^^^^  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/executor/multiproc\_executor.py”, line 107, in **init**  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] super(). **init** (vllm\_config)  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/tracing/otel.py”, line 178, in sync\_wrapper  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] return func(\*args, \*\*kwargs)  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] ^^^^^^^^^^^^^^^^^^^^^  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/executor/abstract.py”, line 109, in **init**  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] self.\_init\_executor()  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/executor/multiproc\_executor.py”, line 200, in \_init\_executor  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] self.workers = WorkerProc.wait\_for\_ready(unready\_workers)  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/executor/multiproc\_executor.py”, line 747, in wait\_for\_ready  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] raise e from None  
(EngineCore pid=122543) ERROR 05-31 14:55:10 [core.py:1140] Exception: WorkerProc initialization failed due to an exception in a background process. See stack trace for root cause.  
(EngineCore pid=122543) Process EngineCore:  
(EngineCore pid=122543) Traceback (most recent call last):  
(EngineCore pid=122543) File “/usr/lib/python3.12/multiprocessing/process.py”, line 314, in \_bootstrap  
(EngineCore pid=122543) self.run()  
(EngineCore pid=122543) File “/usr/lib/python3.12/multiprocessing/process.py”, line 108, in run  
(EngineCore pid=122543) self.\_target(\*self.\_args, \*\*self.\_kwargs)  
(EngineCore pid=122543) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/engine/core.py”, line 1144, in run\_engine\_core  
(EngineCore pid=122543) raise e  
(EngineCore pid=122543) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/engine/core.py”, line 1114, in run\_engine\_core  
(EngineCore pid=122543) engine\_core = EngineCoreProc(\*args, engine\_index=dp\_rank, \*\*kwargs)  
(EngineCore pid=122543) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^  
(EngineCore pid=122543) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/tracing/otel.py”, line 178, in sync\_wrapper  
(EngineCore pid=122543) return func(\*args, \*\*kwargs)  
(EngineCore pid=122543) ^^^^^^^^^^^^^^^^^^^^^  
(EngineCore pid=122543) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/engine/core.py”, line 880, in **init**  
(EngineCore pid=122543) super(). **init** (  
(EngineCore pid=122543) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/engine/core.py”, line 118, in **init**  
(EngineCore pid=122543) self.model\_executor = executor\_class(vllm\_config)  
(EngineCore pid=122543) ^^^^^^^^^^^^^^^^^^^^^^^^^^^  
(EngineCore pid=122543) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/executor/multiproc\_executor.py”, line 107, in **init**  
(EngineCore pid=122543) super(). **init** (vllm\_config)  
(EngineCore pid=122543) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/tracing/otel.py”, line 178, in sync\_wrapper  
(EngineCore pid=122543) return func(\*args, \*\*kwargs)  
(EngineCore pid=122543) ^^^^^^^^^^^^^^^^^^^^^  
(EngineCore pid=122543) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/executor/abstract.py”, line 109, in **init**  
(EngineCore pid=122543) self.\_init\_executor()  
(EngineCore pid=122543) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/executor/multiproc\_executor.py”, line 200, in \_init\_executor  
(EngineCore pid=122543) self.workers = WorkerProc.wait\_for\_ready(unready\_workers)  
(EngineCore pid=122543) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^  
(EngineCore pid=122543) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/executor/multiproc\_executor.py”, line 747, in wait\_for\_ready  
(EngineCore pid=122543) raise e from None  
(EngineCore pid=122543) Exception: WorkerProc initialization failed due to an exception in a background process. See stack trace for root cause.  
(APIServer pid=122183) Traceback (most recent call last):  
(APIServer pid=122183) File “/home/aiuser/llm\_workspace/venv/bin/vllm”, line 10, in  
(APIServer pid=122183) sys.exit(main())  
(APIServer pid=122183) ^^^^^^  
(APIServer pid=122183) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/entrypoints/cli/main.py”, line 92, in main  
(APIServer pid=122183) args.dispatch\_function(args)  
(APIServer pid=122183) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/entrypoints/cli/serve.py”, line 122, in cmd  
(APIServer pid=122183) uvloop.run(run\_server(args))  
(APIServer pid=122183) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/uvloop/ **init**.py”, line 96, in run  
(APIServer pid=122183) return \_\_asyncio.run(  
(APIServer pid=122183) ^^^^^^^^^^^^^^  
(APIServer pid=122183) File “/usr/lib/python3.12/asyncio/runners.py”, line 194, in run  
(APIServer pid=122183) return runner.run(main)  
(APIServer pid=122183) ^^^^^^^^^^^^^^^^  
(APIServer pid=122183) File “/usr/lib/python3.12/asyncio/runners.py”, line 118, in run  
(APIServer pid=122183) return self.\_loop.run\_until\_complete(task)  
(APIServer pid=122183) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^  
(APIServer pid=122183) File “uvloop/loop.pyx”, line 1518, in uvloop.loop.Loop.run\_until\_complete  
(APIServer pid=122183) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/uvloop/ **init**.py”, line 48, in wrapper  
(APIServer pid=122183) return await main  
(APIServer pid=122183) ^^^^^^^^^^  
(APIServer pid=122183) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api\_server.py”, line 693, in run\_server  
(APIServer pid=122183) await run\_server\_worker(listen\_address, sock, args, \*\*uvicorn\_kwargs)  
(APIServer pid=122183) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api\_server.py”, line 707, in run\_server\_worker  
(APIServer pid=122183) async with build\_async\_engine\_client(  
(APIServer pid=122183) File “/usr/lib/python3.12/contextlib.py”, line 210, in **aenter**  
(APIServer pid=122183) return await anext(self.gen)  
(APIServer pid=122183) ^^^^^^^^^^^^^^^^^^^^^  
(APIServer pid=122183) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api\_server.py”, line 100, in build\_async\_engine\_client  
(APIServer pid=122183) async with build\_async\_engine\_client\_from\_engine\_args(  
(APIServer pid=122183) File “/usr/lib/python3.12/contextlib.py”, line 210, in **aenter**  
(APIServer pid=122183) return await anext(self.gen)  
(APIServer pid=122183) ^^^^^^^^^^^^^^^^^^^^^  
(APIServer pid=122183) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/entrypoints/openai/api\_server.py”, line 136, in build\_async\_engine\_client\_from\_engine\_args  
(APIServer pid=122183) async\_llm = AsyncLLM.from\_vllm\_config(  
(APIServer pid=122183) ^^^^^^^^^^^^^^^^^^^^^^^^^^  
(APIServer pid=122183) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/engine/async\_llm.py”, line 217, in from\_vllm\_config  
(APIServer pid=122183) return cls(  
(APIServer pid=122183) ^^^^  
(APIServer pid=122183) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/engine/async\_llm.py”, line 146, in **init**  
(APIServer pid=122183) self.engine\_core = EngineCoreClient.make\_async\_mp\_client(  
(APIServer pid=122183) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^  
(APIServer pid=122183) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/tracing/otel.py”, line 178, in sync\_wrapper  
(APIServer pid=122183) return func(\*args, \*\*kwargs)  
(APIServer pid=122183) ^^^^^^^^^^^^^^^^^^^^^  
(APIServer pid=122183) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/engine/core\_client.py”, line 130, in make\_async\_mp\_client  
(APIServer pid=122183) return AsyncMPClient(\*client\_args)  
(APIServer pid=122183) ^^^^^^^^^^^^^^^^^^^^^^^^^^^  
(APIServer pid=122183) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/tracing/otel.py”, line 178, in sync\_wrapper  
(APIServer pid=122183) return func(\*args, \*\*kwargs)  
(APIServer pid=122183) ^^^^^^^^^^^^^^^^^^^^^  
(APIServer pid=122183) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/engine/core\_client.py”, line 900, in **init**  
(APIServer pid=122183) super(). **init** (  
(APIServer pid=122183) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/engine/core\_client.py”, line 535, in **init**  
(APIServer pid=122183) with launch\_core\_engines(  
(APIServer pid=122183) File “/usr/lib/python3.12/contextlib.py”, line 144, in **exit**  
(APIServer pid=122183) next(self.gen)  
(APIServer pid=122183) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/engine/utils.py”, line 1128, in launch\_core\_engines  
(APIServer pid=122183) wait\_for\_engine\_startup(  
(APIServer pid=122183) File “/home/aiuser/llm\_workspace/venv/lib/python3.12/site-packages/vllm/v1/engine/utils.py”, line 1187, in

---

<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:** [June 5, 2026, 12:52am UTC](https://discuss.vllm.ai/t/runtimeerror-cuda-driver-initialization-failed-you-might-not-have-a-cuda-gpu/2698/2 "2026-06-05T00:52:40Z")

</div>

你的报错“RuntimeError: CUDA driver initialization failed, you might not have a CUDA gpu.”说明PyTorch无法检测到CUDA GPU。根据vLLM官方文档和issue，这通常是环境配置或驱动问题导致的。请确认以下几点：

1. 检查nvidia-smi能否在你的虚拟环境下正常运行，且能看到H800显卡；2. 确认你的PyTorch和CUDA驱动版本完全兼容，vLLM 0.21.0官方推荐CUDA 12.9/13.0和对应PyTorch（如2.11.0+cu130），但驱动需为580.105.08及以上，且必须用官方pip/uv wheel安装，不建议用conda安装PyTorch，否则NCCL等库可能冲突；3. 若用Docker，需加–gpus all参数，并确保容器内nvidia-smi可用；4. 若用裸机，建议新建干净的venv/uv环境，重新pip安装vllm和torch，避免老环境残留影响；5. 若驱动升级后仍报错，可尝试重启服务器或重装驱动。

详细排查步骤和环境变量设置可参考：[vLLM官方安装文档](https://docs.vllm.ai/en/latest/getting_started/installation/gpu.html) 和 [troubleshooting指南](https://docs.vllm.ai/en/latest/usage/troubleshooting/#cuda-error-the-provided-ptx-was-compiled-with-an-unsupported-toolchain)。你需要确保所有依赖都在同一CUDA版本下编译和运行。

需要更详细的排查步骤吗？

Sources:

- [https://docs.vllm.ai/en/latest/usage/troubleshooting/#cuda-error-the-provided-ptx-was-compiled-with-an-unsupported-toolchain](https://docs.vllm.ai/en/latest/usage/troubleshooting/#cuda-error-the-provided-ptx-was-compiled-with-an-unsupported-toolchain)
- [vLLM官方安装文档](https://docs.vllm.ai/en/latest/getting_started/installation/gpu.html)

Hint: Mention @RunLLM in the post for followups.