0

I need some help for setting up my linux box for scaled-yolov4. My linux box have a gpu card, Nvidia quadro P620 2GB card. running linux mint 20.4 = ubuntu 20.4.

I have anaconda setup. But didn't help much. This is what I did.

  1. I installed the cuda 11.2 by using the following:
sudo mv cuda-ubuntu2004.pin /etc/apt/preferences.d/cuda-repository-pin-600
wget https://developer.download.nvidia.com/compute/cuda/11.2.1/local_installers/cuda-repo-ubuntu2004-11-2-local_11.2.1-460.32.03-1_amd64.deb
sudo dpkg -i cuda-repo-ubuntu2004-11-2-local_11.2.1-460.32.03-1_amd64.deb
sudo apt-key add /var/cuda-repo-ubuntu2004-11-2-local/7fa2af80.pub
sudo apt-get update
sudo apt-get -y install cuda

and then using conda to install the rest. In this case, the detect.py works fine, but train.py didn't work with out of memory error.

  1. I installed the CUDA10.2, things work without GPU, so its very slow.

Can someone tell me the right way to install NVIDIA card and CUDA and setup the environment for the scaled-yolov4? Thank you very much.

talonmies
  • 70,661
  • 34
  • 192
  • 269
Ken
  • 11
  • 3
  • 1
    If `detect.py` worked correctly, most likely your environment was setup correctly, and its quite possible that the `train.py` you are using simply requires more than the 2GB that is available on your GPU. – Robert Crovella Feb 23 '21 at 14:26
  • No, not that easy, the train.py always got the wrong size of memory, and state "out of memory", so can't train in the envirnment. – Ken Feb 23 '21 at 14:30
  • You don't have an installation problem. Your assumption that the very low-end mobile GPU is suitable for what you are trying to use it for is wrong – talonmies Feb 23 '21 at 14:36
  • ok, the GPU is not the best, I understand that. So, it should be slow, but it should work, right? but it's not working in train.py. That's the problem. – Ken Feb 23 '21 at 14:54
  • Speed is one thing. Memory is another. It is very possible that 2Gb is insufficient to run the code you are testing – talonmies Feb 23 '21 at 14:59
  • I put the size of image and input to a min, it's just 100 images, in 416 widthxhight. it's very small. 2gb should be fine. and the batch size was 16 or lower, the out of memory was wrong. – Ken Feb 23 '21 at 15:06
  • hi @RobertCrovella , what you guys will do to setup the env? I will try your way to see if I got a better result. Thank you. – Ken Feb 23 '21 at 15:10
  • hi @talonmies what you guys will do to setup the env? I will try your way to see if I got a better result. Thank you. – Ken Feb 23 '21 at 15:11
  • These machine learning frameworks are extremely memory greedy. There is nothing to do at a GPU or operating system configuration level which will change anything: The only problem here is your expectations – talonmies Feb 23 '21 at 22:45
  • Try doing it in Google Colab with their GPU enabled – Addie Ira B. Parico Feb 25 '21 at 09:20

0 Answers0