Error when install CUDA toolkit on Ubuntu 16.04 Virtual Machine

I am installing JetPack 3.1 for TX2 on my virtual machine host but facing error with CUDA toolkit.
Error log:
The following packages have unmet dependencies:
cuda-toolkit-8-0 : Depends: cuda-samples-8-0 (>= 8.0.64) but it is not going to be installed
Depends: cuda-documentation-8-0 (>= 8.0.64) but it is not going to be installed
E: Unable to correct problems, you have held broken packages.
My VM Ubuntu version: Linux osboxes 4.10.0-28-generic #32~16.04.2-Ubuntu SMP Thu Jul 20 10:19:48 UTC 2017 x86_64 x86_64 x86_64 GNU/Linux
Please help.

VMs are usually trouble and don’t work well. I do not know if CUDA will even work on a VM since it needs direct hardware access to the GPU (perhaps a VM has a virtual GPU and not a real one? I don’t know).

Hi dungta28888.ansv, installing JetPack in virtual machine isn’t officially supported, if possible please natively dual-boot Ubuntu 16.04 x86_64 with the host. Even if you were to get CUDA installed in the VM, flashing L4T over USB to TX2, from within a VM, is unreliable.

Hi all, so what is minimum system requirements to install JetPack (for host computer)?

Spare disk space after actually installing JetPack and/or driver package plus sample rootfs is about 40GB of extra space. File system type should be anything native to Linux, e.g., ext4 is the default in most distributions and this is fine. Running on VFAT or NTFS will cause failure.

The recommendation, and requirement if using JetPack, is Ubuntu 14.04 or 16.04 (there are some PC side demos which can’t be installed on 16.04, but which work in 14.04).

Command line (driver package plus sample rootfs) works from any x86_64 PC Linux.

In all cases flash is over USB. If extra packages are installed via JetPack, then you want wired ethernet on the Jetson. Internet access on the PC can be WiFi or wired.

Many people dual boot.

Note that CUDA software requires a host to have an NVIDIA GPU (video card). I have a dual core Atom laptop which works except for the PC host side installs because it doesn’t have the GPU.

@linuxdev: My PC has no NVIDIA GPU too.
Do I have to install CUDA on host side?

Normally CUDA is not required on the host. I suspect the testing release (JetPack3.2 pre-release) does (I have a laptop without an NVIDIA GPU which works with JetPack3.1, but fails in a way that makes me think it is a bug in the installer not working with the Jetson when the host side packages are disabled). Simply uncheck package installs to host when you get to that step.

Hi all, anyone know where to download CUDA toolkit 8.0.82 for Ubuntu? Or any fixes to download CUDA toolkit from JetPack installation wizard?
I have search on Nvidia website but only found 8.0.61 version (CUDA Toolkit 8.0 GA2 (Feb 2017)).

JetPack is the official method of installing CUDA. Is JetPack failing? JetPack3.2 is a pre-release with some newer versions on it, but you may find Jetsons lag slightly behind relative to PC versions. I could be wrong, but I think version 8 is the most recent version on a TX2.

@linuxdev: yes, it failed with message in the very first post on this thread.
Give the error log again:
Error log:
The following packages have unmet dependencies:
cuda-toolkit-8-0 : Depends: cuda-samples-8-0 (>= 8.0.64) but it is not going to be installed
Depends: cuda-documentation-8-0 (>= 8.0.64) but it is not going to be installed
E: Unable to correct problems, you have held broken packages.

Someone knowing more about JetPack (especially 3.2) will need to answer, but it seems odd that packages would exist in JetPack, yet refuse to install…unless some incorrect configuration is chosen. The trouble is that I don’t know what to check. Do be aware though that you can re-run package installs at any time, it won’t hurt anything to try again so long as you don’t pick the flash step.

One possible reason for a package install to fail is if the host does not have the “ssh-askpass” package…this is a case where the pop-up for asking a password won’t exist.

@linuxdev: I managed to install jetpack on target, except cuda samples. It requires host installation, so that I am facing the old problem :).
Thanks for your support.

With R28.1 (JetPack3.1) I don’t have errors. I do not choose to compile samples because this is done on the host PC…and when using JetPack I have to use a dual core 64-bit Atom with no NVIDIA GPU. Think of it as an advanced mechanical calculator and a rubber band to power it :P

You can install source code for samples without compiling to executable. Because these are cross compiled it will demand putting the samples on the host if you choose compile.

@linuxdev: you mean I can install source code for samples on the target? If so, how to do it?
Sorry if noob question but I am new to Jetson.

JetPack has software it can install to the host as well as to the Jetson. The host side packages are aids to supporting development of the Jetson…cross compiler, debugger, so on. When you install components to the host you will find them in a subdirectory of “/usr/local”. Sample source code for CUDA programming will be at:

/usr/local/cuda-8.0/samples

(in the case of a future CUDA version upgrade it would be a similar path)

This sample code cannot run on the host itself unless it has an NVIDIA video card with NVIDIA driver (Nouveau driver will not work with CUDA).