PaddlePaddle / PaddlePaddle/FastDeploy
自行编译的FastDeploy,部署NVIDIA Gpu PPOCR时只有cuda11.2版本能运行,但仍然有报错
Open
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 3.7k
- Forks
- 756
- Avg merge
- 19h 28m
- Merged PRs (30d)
- 4
Description
环境
- 【编译命令】:自行编译的FastDeploy,编译方式使用vs2019(x64 Native Tools Command Prompt for VS 2019)
GPU版
cmake .. -G "Visual Studio 16 2019" -A x64 -DENABLE_PADDLE_BACKEND=ON
-DENABLE_VISION=ON
-DENABLE_TEXT=ON
-DWITH_GPU=ON
-DCUDA_DIRECTORY="D:\cuda"
-DCMAKE_INSTALL_PREFIX="D:\Paddle\compiled_fastdeploy_cuda11.2"
msbuild fastdeploy.sln /m /p:Configuration=Release /p:Platform=x64
msbuild INSTALL.vcxproj /m /p:Configuration=Release /p:Platform=x64
CPU版
cmake .. -G "Visual Studio 16 2019" -A x64 -DENABLE_OPENVINO_BACKEND=ON
-DENABLE_VISION=ON
-DENABLE_TEXT=ON
-DCMAKE_INSTALL_PREFIX="D:\Paddle\compiled_fastdeploy_cpu"
msbuild fastdeploy.sln /m /p:Configuration=Release /p:Platform=x64
msbuild INSTALL.vcxproj /m /p:Configuration=Release /p:Platform=x64 - 【系统平台】:Windows10 x64 Intel(R) Xeon(R) Silver 4110 CPU
- 【硬件】:Nvidia GPU RTX 2070, CUDA 11.2.0, CUDNN 8.2.0 ,最高cuda版本支持12.2
- 【编译语言】: C++
推理
- 【推理代码】
https://github.com/PaddlePaddle/FastDeploy/blob/develop/examples/vision/ocr/PP-OCR/cpu-gpu/cpp/infer.cc
推理模型使用的是ch_PP-OCRv3系列 - 【CPU版推理】:
单张图片推理结果没有问题,使用opencvVINO推理框架,输入尺寸640*480,循环跑速度200ms左右 - 【GPU版推理】:
单张图片推理结果没有问题,但是有个CUDA释放错误(乱码是cmd编码问题,不影响CUDA报错),使用飞浆推理框架,输入尺寸640*480,循环跑速度140ms左右
- 【尝试解决】
用的是官方的代码,代码肯定没有问题,尝试其他cuda版本(cuda11.5.2,cudnn8.3.3)(cuda11.6.0,cudnn8.4.0),在运行demo时都出现另一个错误,程序跑不起来,忘了截图
Could not load library cudnn_cnn_infer64_8.dll.
我把cudnn_cnn_infer64_8.dll放到当前目录了不行,下载了zlibwapi.dll放在exe当前目录和配置环境变量不行,保持cuda版本不动使用了几个其他版本的cudnn_cnn_infer64_8.dll不行
问题
- 【问题请教】
三个问题请教大佬,折腾好几天没有解决
1.(cuda11.2.0,cudnn8.2.0)的报错CUDA error(4), driver shutting down有什么思路吗
2.(cuda11.5.2,cudnn8.3.3)(cuda11.6.0,cudnn8.4.0)的报错Could not load library cudnn_cnn_infer64_8.dll.有什么思路吗
3.GPU版demo发送到另外一台机子跑,速度变成了500ms,他的显卡是1070,是显卡的原因吗
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start with examples/vision/ocr/PP-OCR/cpu-gpu/cpp/infer.cc and reproduce the reported Windows GPU runs using the listed CUDA and cuDNN combinations. Check the CUDA error, missing cudnn_cnn_infer64_8.dll failure, and performance difference on the RTX 2070 and GTX 1070; done means identifying the cause or documenting the supported configuration and required runtime libraries.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- cpp
- Domain
- machine-learning
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100