No GPG keys error when installing CUDA6.5 on Ubuntu 14.04 x84_64

Hi,

I already submitted the query below to the CUDA setup and install section. But I wasn’t getting any reply and I was told that I should re-post in Embedded systems:

I recently bought the Jetson TK1 board and I managed to install CUDA6.5 on it to do native compilation. However, I also want to do cross compilation. To that end, I am trying to do a cross-build installation on my Ubuntu 14.04 x86 64 bit host machine. More precisely, I am trying to perform the cross-build installation on a Ubuntu guest virtualbox I have running on my Windows host.

I would like to add that the GPU on my host machine is not CUDA capable. But I don’t think this matters because I only intend to compile CUDA programs on the host machine, not run them.

The steps I am following are in section 5.1 of the CUDA setup and installation guide: https://devtalk.nvidia.com/default/topic/803524/embedded-systems/anyone-have-a-guide-for-cloning-jetson-tk1-without-the-jetpack-development-pack-/

When I try to run sudo dpkg -i cuda-repo-ubuntu1404_6.5-14_amd64.deb, I get the following error:

auggy@auggy-VirtualBox:~/Downloads$ sudo dpkg -i cuda-repo-ubuntu1404_6.5-14_amd64.deb
Selecting previously unselected package cuda-repo-ubuntu1404.
(Reading database … 163905 files and directories currently installed.)
Preparing to unpack cuda-repo-ubuntu1404_6.5-14_amd64.deb …
Unpacking cuda-repo-ubuntu1404 (6.5-14) …
Setting up cuda-repo-ubuntu1404 (6.5-14) …
gpg: no valid OpenPGP data found.
Failed to add GPGKEY at http://cuda-repo/repos/GPGKEY to apt keys.

I am pretty sure I am using the correct operating system:

auggy@auggy-VirtualBox:~/Downloads$ uname -m && cat /etc/*release
x86_64
DISTRIB_ID=Ubuntu
DISTRIB_RELEASE=14.04
DISTRIB_CODENAME=trusty
DISTRIB_DESCRIPTION=“Ubuntu 14.04.2 LTS”
NAME=“Ubuntu”
VERSION=“14.04.2 LTS, Trusty Tahr”
ID=ubuntu
ID_LIKE=debian
PRETTY_NAME=“Ubuntu 14.04.2 LTS”
VERSION_ID=“14.04”
HOME_URL=“http://www.ubuntu.com/
SUPPORT_URL=“http://help.ubuntu.com/
BUG_REPORT_URL=“http://bugs.launchpad.net/ubuntu/

I would really appreciate some help on this issue. I am fairly new to Ubuntu so please bear with me.

Thanks!

Apologies, the link to the CUDA setup and install guide is actually http://developer.download.nvidia.com/compute/cuda/6_5/rel/docs/CUDA_Getting_Started_Linux.pdf.

The quoted guide URL seems to have a bug in it. I get access denied even though I’m logged in correctly if I directly click on the link. The general URL for CUDA 6.5 is here, providing indirect access to the guide:
https://developer.nvidia.com/cuda-toolkit-65

Sadly, the main documentation URL of the 6.5 page is a circular reference back to 7.0 docs, with the main 6.5 docs not available…the 6.5 page documentation link should probably point to the 6.5 docs, not the 7.0 docs (then again, this is a different error than the thread topic).

So the 6.5 guide simply can’t be downloaded without first navigating to the 6.5 CUDA page and clicking the linux tab (then searching for “Linux Getting Started Guide”). The direct link is context-broken (a design pattern derived from “context-sensitive” :). Clicking the guide link from the 6.5 page works, clicking the guide link from elsewhere denies access.

Unfortunately I only have a Fedora x86 host, and cannot test the cuda-repo-ubuntu deb. It’s kind of a wild guess, but it is possible that the GPG signature part has a similar web page access issue and is simply denying the GPG URL during the attempt to install the repo definition package. Regardless of WHY the GPG signature is not working, I would try the dpkg option “–no-debsig”.

Hi,

I tried the --no-debsig option but it did not work:

auggy@auggy-VirtualBox:~/Jetson/cuda$ sudo dpkg -i --no-debsig cuda-repo-ubuntu1404_6.5-14_amd64.deb
Selecting previously unselected package cuda-repo-ubuntu1404.
(Reading database … 163905 files and directories currently installed.)
Preparing to unpack cuda-repo-ubuntu1404_6.5-14_amd64.deb …
Unpacking cuda-repo-ubuntu1404 (6.5-14) …
Setting up cuda-repo-ubuntu1404 (6.5-14) …
gpg: no valid OpenPGP data found.
Failed to add GPGKEY at http://cuda-repo/repos/GPGKEY to apt keys.

Can I install CUDA7 if my aim is to just do cross compilation for my Jetson which has CUDA6.5?

For Jetson’s L4T version R19.x, only CUDA 6.0 works. For Jetson’s L4T version R21.x, only CUDA 6.5 works. It has been stated that it is unlikely CUDA 7 will be ported to any 32 bit platform. For that you’ll probably need to wait for Tegra X1.

I’ll drop a note to someone that the GPGKEY is failing. FYI, I cannot get DNS to resolve anything with cuda-repo (http://cuda-repo isn’t even a real domain), so the site has either changed or something is cutting it off.

Sorry, I meant to ask if I could install CUDA 7.0 on my x86 host machine for cross compilation with my Jetson. I am aware that Jetson only supports up to CUDA 6.5.

Thanks for looking into this issue!

You can install 7 on the desktop. Nothing it builds will run on Jetson though.

Alright, then I guess I need to wait until Nvidia fixes this issue?

I see the same GPG error with the cuda-repo package for Ubuntu 14.04. The package seems to be plain broken.

I think as a workaround, you can install the cuda-repo package for 12.04 as that seems to work. Then edit /etc/apt/sources.list.d/cuda.list and change 12.04 to 14.04 and run “apt-get update”.

I tried this workaround and I end up installing CUDA 7.0, not CUDA 6.5. Here are the steps I performed:

  1. Download cuda-repo-ubuntu1204_6.5-14_amd64.deb.
  2. Because I want to do cross-compilation, I enabled armhf as a foreign architecture with the command “sudo dpkg --add-architecture armhf”
  3. sudo apt-get update
  4. sudo dpkg -i cuda-repo-ubuntu1204_6.5-14_amd64.deb
  5. Edited /etc/apt/sources.list.d/cuda.list as per the workaround.

The original cuda.list file had the following:

deb http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1204/x86_64 /

I changed it to:

deb http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1404/x86_64 /

  1. sudo apt-get update
  2. sudo apt-get install cuda-cross-armhf

Hi,

Is it okay if I just got a Ubuntu 12.04 x86 64 bit machine and just intall CUDA 6.5 onto that? It’s a bit troublesome but that’s probably the most straight-forward solution.

Are you suggesting that because you want 6.5 and with the above you got 7.0?

The repositories seem to provide both. If you do “apt-cache search cuda”, you’ll see a long list of packages, some with version numbers, some without. The packages without version number in the package name typically depend on the latest version.

So try this:

sudo apt-get purge cuda-cross-armhf 
sudo apt-get install cuda-cross-armhf-6-5

Thanks, that worked!