maxmustermann/nvidia-docker
1
0
Fork 0
Code Issues Pull requests Projects Releases Wiki Activity
396 commits 5 branches 21 tags 20 MiB
Commit graph

396 commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
Felix Abecassis df28bb4600 Fix release number for the rpm packages 2017-09-28 15:48:01 -07:00
Christy Perez 09fc5d78f1 Add rpm support for ppc64le 
Signed-off-by: Christy Perez <christy@linux.vnet.ibm.com>
Signed-off-by: Felix Abecassis <fabecassis@nvidia.com>
 2017-09-18 14:08:11 -07:00
Christy Perez f00167fe3f Add ppc64le support for "make deb" 
Signed-off-by: Christy Perez <christy@linux.vnet.ibm.com>
Signed-off-by: Felix Abecassis <fabecassis@nvidia.com>
 2017-09-18 13:15:05 -07:00
Felix Abecassis 9deb68f707 Improve the portability of the debian/control file 2017-09-06 19:24:11 -07:00
Felix Abecassis bbdc9dc342 Add support for 17.06.2 2017-09-06 15:03:54 -07:00
Felix Abecassis 376a0efd0e Initial commit for branch 2.0 2017-09-05 19:16:49 -07:00
Felix Abecassis d8e8727788 Update README.md 2017-07-22 12:54:03 -07:00
Renaud Gaubert cac5e9ffad Fix infinite loop when no devices are found 
Signed-off-by: Renaud Gaubert <rgaubert@nvidia.com>
Signed-off-by: Felix Abecassis <fabecassis@nvidia.com>
 2017-05-26 10:28:02 -07:00
Felix Abecassis 62ad007333 Fix out-of-bounds access with "nvidia-docker container" 2017-05-24 16:24:29 -07:00
Jonathan Calmels 01d2c94366 Update changelog for 1.0.1 2017-03-02 18:15:33 -08:00
Felix Abecassis 16fd00ac13 Support new Docker package names 
RPM packages do not support multiple options for one dependency, and
the new packages don't have a "Provides: docker-engine", so our only
option was to remove the dependency on Docker.
 2017-03-02 14:33:04 -08:00
Felix Abecassis 16a7d7da64 Unconditionally call nvidia-modprobe to load UVM 
Checking the existence of /dev/nvidia-uvm had two limitations:
- If kernel module nvidia_uvm is unloaded, /dev/nvidia-uvm is not
  removed, so nvidia-modprobe was not called when it should have.
- Even if kernel module nvidia_uvm is loaded and /dev/nvidia-uvm
  exists, /dev/nvidia-uvm-tools might not exist. Using nvidia-modprobe
  fixes this issue.

nvidia-modprobe parses /proc/modules to check which modules are
already loaded, so there should be no impact on performance.
 2017-02-27 16:40:08 -08:00
Felix Abecassis 7dbc2cafec Attempt to be compatible with newer versions of Docker 
Previously, nvidia-docker aborted if it detected a version of Docker
more recent than the one it supports. In hindsight, this provided
little benefits since the list of flags has now become quite stable,
and it forced our users to frequently update nvidia-docker manually,
because nvidia-docker is not available on a public package repository.

This problem is exacerbated now that Docker is on a monthly release schedule:
https://github.com/docker/docker/pull/31075
We don't want to generate new packages every month.

Thus, if the version of Docker detected by nvidia-docker is newer than
its latest known Docker version, we will use the boolean flags of this
latest known version.
 2017-02-27 15:16:42 -08:00
Felix Abecassis 3c3ddb4919 Add support for "docker container run" 2017-02-27 14:34:05 -08:00
Felix Abecassis 94377628bd Add support for boolean option --init 
Closes: #295
 2017-02-27 13:56:52 -08:00
Felix Abecassis ec285e6b55 Remove Docker version check in the Makefile 
Old Docker versions are less likely to be used now, and with the new
versioning used in Docker, it's not worth the hassle to maintain this
check.
 2017-02-27 13:56:52 -08:00
Felix Abecassis 25a18f513b Print log of image pull to stderr 
Closes: #310
 2017-02-27 13:56:45 -08:00
Felix Abecassis 3422bca272 Move sources to the top-level of the repository 2017-02-06 13:39:25 -08:00
Felix Abecassis ffcf3becde Remove all Dockerfiles, they are now hosted on GitLab 
We now use the public GitLab CI to automatically build images and push
them to DockerHub.

CUDA: https://gitlab.com/nvidia/cuda
Caffe: https://gitlab.com/nvidia/caffe
DIGITS: https://gitlab.com/nvidia/digits
OpenCL: https://gitlab.com/nvidia/opencl
 2017-02-06 11:39:13 -08:00
Felix Abecassis 4ff890ba18 Make samples use the prebuilt images on DockerHub 2017-02-06 10:56:30 -08:00
Jonathan Calmels fbf1e42879 Update changelog for 1.0.0 
Closes: #292
 2017-01-18 14:10:33 -08:00
Felix Abecassis d8f28b6b16 Add missing apt-get option "--no-install-recommends" in a few Dockerfiles 2017-01-11 15:18:55 -08:00
Felix Abecassis 677434f1b7 Remove unused folder symlinks 2017-01-11 15:18:48 -08:00
Felix Abecassis ad5d6b5f5d Prune the list of CUDA tags 
Closes: #268
 2017-01-11 15:18:40 -08:00
Felix Abecassis 94760f38f7 Don't add an empty entry to LD_LIBRARY_PATH 
LD_LIBRARY_PATH is never set in the base images we use.
 2016-12-16 14:22:36 -08:00
Felix Abecassis 0b4480b6ee Remove reference to non-existing path /usr/local/cuda/lib 
This path was for 32-bit installations of CUDA
 2016-12-16 14:22:31 -08:00
Felix Abecassis e0aae98a09 Fix double call to r.body.Close() on the same object 
Partial fix for #123.
 2016-12-06 16:08:51 -08:00
Felix Abecassis 703b71775f Return NUMA node 0 if NUMA support isn't enabled 
Closes: #198
 2016-12-06 16:02:10 -08:00
Felix Abecassis 92c6cd856f Add a requirement on setcap(8) 
Closes: #239
 2016-12-06 16:02:02 -08:00
Christy Perez 5989fe40ce Fix docker run command for systems with selinux 
Add the :Z option on the volume-mount

Closes: #203
Signed-off-by: Christy Perez <christy@linux.vnet.ibm.com>
Signed-off-by: Felix Abecassis <fabecassis@nvidia.com>
 2016-12-05 11:28:15 -08:00
Felix Abecassis ef3e605505 Remove deprecated instruction MAINTAINER 
Closes: #260
 2016-12-02 10:26:27 -08:00
Felix Abecassis 00ebd530fd Fix location of the config file for DIGITS 4.0 
Closes: #252
 2016-11-23 11:41:38 -08:00
Felix Abecassis dae9950caa Add support for plugin endpoint VolumeDriver.Capabilities 2016-11-15 16:11:59 -08:00
Felix Abecassis 10f7ea1ac3 Add support for Docker 1.13 2016-11-15 16:11:59 -08:00
Felix Abecassis 369c559c9d Remove useless quotes in format template 2016-11-15 16:11:52 -08:00
Felix Abecassis 4a5c3d2f05 Add CUDA 8.0 on CentOS 6 
Closes: #227
 2016-10-24 13:04:37 -07:00
Felix Abecassis 8bbdd3ec91 Update cuDNN v4 on Ubuntu 14.04 to 4.0.8 
Closes: #213
 2016-10-10 15:17:14 -07:00
Felix Abecassis a7dcee135e Add missing cuda.repo file for CUDA 8.0 on CentOS 7 2016-10-03 15:43:20 -07:00
Felix Abecassis 0935713b86 Fix repository of CUDA 8.0 on Ubuntu 16.04 
Closes: #212
 2016-10-03 07:53:09 -07:00
Felix Abecassis 81afc02d45 Update CUDA 8.0 from RC to GA 2016-09-30 18:18:41 -07:00
Felix Abecassis 75229893d2 Update Caffe version to 0.15.13 2016-09-30 15:53:32 -07:00
Felix Abecassis 86dcd825fd Remove unnecessary "--force-yes" passed to apt-get commands 2016-09-30 15:34:35 -07:00
Felix Abecassis a52d6681e5 Remove unnecessary ENV instructions 
Use simple variables instead, this reduces the number of layers.
 2016-09-29 10:27:53 -07:00
Felix Abecassis 498b62a25a Update the signing keys used for the official CUDA repository 2016-09-28 23:05:42 -07:00
Christy Perez 18b5d4c330 Fix ldcache path lookup for ppc64le 
Closes: #194
Signed-off-by: Christy Perez <christy@linux.vnet.ibm.com>
Signed-off-by: Jonathan Calmels <jcalmels@nvidia.com>
 2016-09-19 10:28:54 -07:00
Felix Abecassis 257f5d61f7 Update DIGITS 4.0 to use the new deb package 2016-08-11 21:41:06 -07:00
Felix Abecassis 3f9f4b88e4 Add NVIDIA/caffe 0.15 based on CUDA 7.5 2016-08-11 18:08:02 -07:00
Felix Abecassis c1166ae720 Fix Caffe 0.14 Dockerfile 
Now that Caffe 0.15 is available, we need to pin the version of libcaffe-nv0.
 2016-08-11 18:07:35 -07:00
Felix Abecassis ba8012363e Update cuDNN to v5.1 (GA) 2016-08-11 18:06:51 -07:00
Felix Abecassis e902b49cb7 Use official CUDA 8.0 RC patch to enable gcc 5.4 support 
Closes: #167
 2016-08-08 22:08:49 -07:00