An open API service indexing awesome lists of open source software.

https://github.com/andyneff/cuda_sample_docker


https://github.com/andyneff/cuda_sample_docker

Last synced: 5 months ago
JSON representation

Awesome Lists containing this project

README

          

# README #

Simple example for a docker with cuda support using docker+

## Usage ##

git clone https://github.com/andyneff/cuda_sample_docker.git .
git submodule init
git submodule update
make
make run

Now that you are inside the docker

cd /usr/local/cuda/samples/1_Utilities/deviceQuery
make
./deviceQuery

## FAQ ###

1. What if I want to build a docker with a different version of the nvidia driver
or cuda library?

Well, be careful on the first part. As far as I know YOU CAN NOT deviate from the
nvidia driver at all, up or down. It's a bad idea to try. So unless you are just
pre-building images to ship, don't do this. For example:

make NVIDIA_VERSION=352.55

Now changing versions of cuda, is much... safer. The rule is: Your nvidia driver
can support up to a specific version of cuda. It will also work with versions older
than that version, but now newer... To make with a specific version of cuda

make CUDA_VERSION=7.0

The only reliable way I know to check what the maximum version of cuda your
nvidia driver will support is to run the deviceQuery sample cuda program, and
look at the final lines. Similar to this:

deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 7.5, CUDA Runtime Version = 5.5, NumDevs = 3, Device0 = Tesla K20c, Device1 = GeForce GTX 580, Device2 = GeForce GTX 680
Result = PASS

This says that while I'm using cuda 5.5 for this test, My nvidia driver
version (352.55 in this case) can support up to cuda 7.5.