[Help] wsl docker 调用 cuda 核心的问题

[OrdinarySF]

Problem Overview

wsl docker gpu version 无法正常转换字幕

Steps Taken

1. 执行docker命令sudo docker run -itd --gpus 'all,"capabilities=compute,utility,video"' --name bilive_docker_gpu -p 22333:2233 ghcr.io/timerring/bilive-gpu:0.2.10
2. 按照流程启动服务
3. 添加录制任务

Expected Outcome

(A clear and concise description of what you expected to happen.)

Screenshots

scan-20250323.log

Environment Information

• Operating System: Windows 11 WSL Ubuntu 24.10
• Python Version: same docker
• Driver & CUDA Version: Driver Version: 572.16 CUDA Version: 12.8
• Error Messages and Logs: [If applicable, provide any error messages or relevant log outputs]

Additional context

(Add any other context about the problem here.)

[timerring]

wsl 跑 nvidia gpu docker，需要 nvidia-container-toolkit 等配置。

参考官方文档：https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html

[OrdinarySF]

安装过

root@ORDINARY:~# nvidia-smi
Mon Mar 24 23:53:28 2025
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 570.86.16              Driver Version: 572.16         CUDA Version: 12.8     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA GeForce RTX 3070 ...    On  |   00000000:01:00.0 Off |                  N/A |
| N/A   56C    P8             16W /  125W |     934MiB /   8192MiB |     26%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|  No running processes found                                                             |
+-----------------------------------------------------------------------------------------+

[timerring]

好的，nvcc -V cuda 版本是多少

[OrdinarySF]

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2025 NVIDIA Corporation
Built on Fri_Feb_21_20:23:50_PST_2025
Cuda compilation tools, release 12.8, V12.8.93
Build cuda_12.8.r12.8/compiler.35583870_0

[timerring]

单独运行一下识别模块，看一下输出结果，python src/subtitle/generate.py path/to/video

[timerring]

或者直接拉个镜像验证一下。

docker run --rm --gpus all nvidia/cuda:12.0.1-runtime-ubuntu22.04 nvidia-smi

[OrdinarySF]

@timerring sorry最近有点忙

docker run --rm --gpus all  nvidia/cuda:12.0.1-runtime-ubuntu22.04 nvidia-smi
Unable to find image 'nvidia/cuda:12.0.1-runtime-ubuntu22.04' locally
12.0.1-runtime-ubuntu22.04: Pulling from nvidia/cuda
aece8493d397: Pull complete
72951eebaee1: Pull complete
6380df36b797: Pull complete
dce5449f8cbe: Pull complete
fea6c9fe8a9b: Pull complete
4a1fbbf21bd0: Pull complete
1b99e878d962: Pull complete
622df319ef96: Pull complete
7c3b140ed2ea: Pull complete
Digest: sha256:912e635ceffa8c2895e1cc693fd1dc5b5b920e178583cb34ed84a903b439601b
Status: Downloaded newer image for nvidia/cuda:12.0.1-runtime-ubuntu22.04

==========
== CUDA ==
==========

CUDA Version 12.0.1

Container image Copyright (c) 2016-2023, NVIDIA CORPORATION & AFFILIATES. All rights reserved.

This container image and its contents are governed by the NVIDIA Deep Learning Container License.
By pulling and using the container, you accept the terms and conditions of this license:
https://developer.nvidia.com/ngc/nvidia-deep-learning-container-license

A copy of this license is made available in this container at /NGC-DL-CONTAINER-LICENSE for your convenience.

Sun Apr 13 15:37:06 2025
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 570.86.16              Driver Version: 572.16         CUDA Version: 12.8     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA GeForce RTX 3070 ...    On  |   00000000:01:00.0 Off |                  N/A |
| N/A   64C    P8             14W /  105W |    1196MiB /   8192MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|  No running processes found                                                             |
+-----------------------------------------------------------------------------------------+

[timerring]

没事，如果驱动能正常使用的话，可以直接在 docker container 里验证一下 torch 是否能正确调用 cuda 核心:

import torch

def check_cuda_with_pytorch():
    """Check if the PyTorch CUDA environment is working correctly"""
    try:
        print("Checking PyTorch CUDA environment:")
        if torch.cuda.is_available():
            print(f"CUDA device is available, the current CUDA version is: {torch.version.cuda}")
            print(f"PyTorch version is: {torch.__version__}")
            print(f"Detected {torch.cuda.device_count()} CUDA devices.")
            for i in range(torch.cuda.device_count()):
                print(f"Device {i}: {torch.cuda.get_device_name(i)}")
                print(f"Device {i} total memory: {torch.cuda.get_device_properties(i).total_memory / (1024 ** 3):.2f} GB")
                print(f"Device {i} current memory usage: {torch.cuda.memory_allocated(i) / (1024 ** 3):.2f} GB")
                print(f"Device {i} max memory usage: {torch.cuda.memory_reserved(i) / (1024 ** 3):.2f} GB")
        else:
            print("CUDA device is not available.")
    except Exception as e:
        print(f"Error when checking PyTorch CUDA environment: {e}")

if __name__ == "__main__":
    check_cuda_with_pytorch()

[OrdinarySF]

看起来不可用

[timerring]

好的，大概率还是 CUDA 的问题，建议检查一下环境，确保 CUDA cores 可以正常被调用。

[OrdinarySF]

重新拉取了最新版镜像，问题解决