Darknet: yolo with cudnn

Created on 8 Aug 2020 · 6Comments · Source: pjreddie/darknet

I install NVIDIA drivers(version is 450.57)，CUDA11.0，cudnn8.0.1，then I modify "GPU=1 CUDNN=1" in Makefile，But when make，a error has occurred：

./src/convolutional_layer.c: In function ‘cudnn_convolutional_setup’:
./src/convolutional_layer.c:148:5: warning: implicit declaration of function ‘cudnnGetConvolutionForwardAlgorithm’; did you mean ‘cudnnGetConvolutionForwardAlgorithm_v7’? [-Wimplicit-function-declaration]
cudnnGetConvolutionForwardAlgorithm(cudnn_handle(),
^~~~~~~~~
cudnnGetConvolutionForwardAlgorithm_v7
./src/convolutional_layer.c:153:13: error: ‘CUDNN_CONVOLUTION_FWD_SPECIFY_WORKSPACE_LIMIT’ undeclared (first use in this function); did you mean ‘CUDNN_CONVOLUTION_FWD_ALGO_DIRECT’?
CUDNN_CONVOLUTION_FWD_SPECIFY_WORKSPACE_LIMIT,
^~~~~~~~~~~
CUDNN_CONVOLUTION_FWD_ALGO_DIRECT
compilation terminated due to -Wfatal-errors.
Makefile:89: recipe for target 'obj/convolutional_layer.o' failed
make: * [obj/convolutional_layer.o] Error 1

What could be the reason for this situation?

Source

chasenzhang5

Most helpful comment

O-key. I did it by the methos at Downgrade CUDA for Tensorflow-GPU

It works!

Michi-123 on 14 Sep 2020

🚀1 ❤1 🎉1

All 6 comments

I have the same issue using NVIDIA drivers(version is 450.57)，CUDA11.0，cudnn11.0.2.39.

RoboNuke on 30 Aug 2020

I have the same issue using NVIDIA drivers(version is 450.57)，CUDA11.0，cudnn11.0.2.39.

I install this, then run yolov3 secceed.
nvidia-driver-440
CUDA10.2
cudnn7.6.5

chasenzhang5 on 30 Aug 2020

👍1

I suspected it was a version issue. Ill give this a try tomorrow. Thank you so much for you help!

RoboNuke on 30 Aug 2020

I install this, then run yolov3 secceed.
nvidia-driver-440
CUDA10.2
cudnn7.6.5

How did you down grade from cuda11.0 to cuda10.2?
I removed cuda by bellow methods.

sudo apt purge cuda*
sudo apt purge nvidia-cuda-*
sudo apt purge *nvidia*
sudo apt purge libcuda*
sudo apt purge *nsight*
sudo apt-get --purge remove nvidia-*
sudo apt-get --purge remove *nvidia*
sudo apt-get --purge remove *nsight*
sudo apt-get --purge remove cuda-*
sudo apt-get --purge remove cuda
sudo apt autoremove

And installed CUDA Toolkit 10.2 Download .

But I got this.

$ cat  /usr/local/cuda/version.txt
CUDA Version 11.0.228

$ ls -ld /usr/local/cuda*
lrwxrwxrwx  1 root root    9  9月 12 18:30 /usr/local/cuda -> cuda-11.0
drwxr-xr-x 15 root root 4096  9月 12 18:28 /usr/local/cuda-11.0

Do you have any idea? Thanks.

Michi-123 on 14 Sep 2020

O-key. I did it by the methos at Downgrade CUDA for Tensorflow-GPU

It works!

Michi-123 on 14 Sep 2020

🚀1 ❤1 🎉1

O-key. I did it by the methos at Downgrade CUDA for Tensorflow-GPU

It works!

Thanks for the recommendation, it helped me to install clean setup. If someone has problems executing the commands in that blog, type them by hand. Sometimes characters is not recognized in terminal when copy and paste.