Notice: Undefined variable: name in /srv/http/vhosts/ on line 248

Package Details: cudnn 5.1.5-1

Git Clone URL: (read-only)
Package Base: cudnn
Description: NVIDIA CUDA Deep Neural Network library
Upstream URL:
Licenses: proprietary
Submitter: bchretien
Maintainer: archdria
Last Packager: archdria
Votes: 23
Popularity: 1.352783
First Submitted: 2015-11-11 02:42
Last Updated: 2016-10-25 12:07

Dependencies (1)

Sources (0)

Latest Comments

1 2 Next › Last »

spider-mario commented on 2016-11-30 22:42

You could skip the prepare() function entirely by using a file:// source, as noted here:

archtumn commented on 2016-10-24 03:26

Updating as @tiagoshibata says would be nice, if not please edit PKGBUILD and it seems to build fine

acgtyrant commented on 2016-10-11 12:16

Yes! Update as tiagoshibata said please!

tiagoshibata commented on 2016-10-01 14:19

CUDA 8 is out of RC and latest version of cuDNN is v5.1 for CUDA 8.0. Could you update it?

Tar filename is cudnn-8.0-linux-x64-v5.1.tgz, SHA256 is a87cb2df2e5e7cc0a05e266734e679ee1a2fadad6f06af82a76ed81a23b102c8.

epitron commented on 2016-07-18 22:29

It's probably a good idea to make the dependency version-specific as well, ie: depends=('cuda>=7.5')

bchretien commented on 2016-07-18 20:00

@epitron: thanks! I just moved to the 5.1 RC (backwards compatible with a few improvements apparently), as a few users suggested it and the changelog seems rather innocent. The docker still relies on the latest stable release (5.0).

epitron commented on 2016-07-18 19:42

I made a couple tweaks to the PKGBUILD that makes it automatically look in the current directory for the package's tarball:

I also switched it to the latest cudnn that I could find in nvidia's Dockerfile:

bchretien commented on 2016-07-17 18:42

Package updated to the 5.1 RC.

bchretien commented on 2016-05-31 10:55

@quyetnd: you can easily make one.

1) Start with this:
2) Rename the package, set it as providing "cudnn=4.0", and either mark it as conflicting with "cudnn", or avoid conflicts by moving the files to a separate non-standard directory (in that case, tensorflow's PKGBUILD would need to be adapted accordingly).

quyetnd commented on 2016-05-31 04:22

Can we have a separate package for cudnn 4 ? Since tensorflow currently only support cudnn v4