Help:Cuda linux
CUDA linux hints
This guide is useful for nvidia cards such as the tegrity or tesla.
For information on driver installation, see nvidia.
Useful commands[edit]
- nvidia-smi
- module load cuda cudnn
- nvidia-ps (locally written -- ask if you don't have it)
- nvcc
NOTE: If you get the following message, do NOT install cuda. First check if /usr/local/cuda/bin exists.
The pprogram 'nvcc' is not installed. You can install it by typing: apt-get install nvidia-cuda-toolkit
DO NOT DO THIS. Most likely cuda is already installed but /usr/local/cuda/bin is not in your path. If this directory does not exist, follow instructions at 'nvidia' to correctly install cuda.
Cuda capable software[edit]
If you have a gpu and your server is missing any of these and you'd like to use them, let us know.
- cudann (nvidia) nVidia deep learning library
- caffe (needs boost opencv) (install)
- theano (via pip)
- torch : dependencies readline-devel gnuplot zeromq-devel nodejs (qt for qutlua, qttorch) (sox for audio)
- managedCUDA (C#)
- TensorFlow / download and setup (not installed yet)
Cuda with SGE[edit]
gpu.sge:
#$ -cwd #$ -l gpu=1 module load cuda opencv caffe caffe.bin train --solver=solver.prototxt
Note that this script contains common options, you should adjust the module list to fit your needs.
RTX issues with tensorflow[edit]
If your job works on pascal gpus but not the new RTX gpus your code may be having memory segmentation issues. Try this:
from tensorflow.compat.v1 import ConfigProto from tensorflow.compat.v1 import InteractiveSession config = ConfigProto() config.gpu_options.allow_growth = True session = InteractiveSession(config=config)